Get ebook downloads in full at ebookname.

com

Multiresolution Signal Decomposition Transforms

Subbands and Wavelets 2nd Edition Akansu

https://ebookname.com/product/multiresolution-signal-
decomposition-transforms-subbands-and-wavelets-2nd-edition-
akansu/

OR CLICK BUTTON

DOWNLOAD EBOOK

Explore and download more ebook at https://ebookname.com

Instant digital products (PDF, ePub, MOBI) available
Download now and explore formats that suit you...

Signal and Image Multiresolution Analysis 1st Edition

Abdelialil Ouahabi

https://ebookname.com/product/signal-and-image-multiresolution-
analysis-1st-edition-abdelialil-ouahabi/

Quaternion Fourier Transforms for Signal and Image

Processing 1st Edition Todd A. Ell

https://ebookname.com/product/quaternion-fourier-transforms-for-
signal-and-image-processing-1st-edition-todd-a-ell/

Wavelet Transforms and Their Applications 2nd edition

Edition Debnath

https://ebookname.com/product/wavelet-transforms-and-their-
applications-2nd-edition-edition-debnath/

The End of the Ancient World Ferdinand Lot

https://ebookname.com/product/the-end-of-the-ancient-world-
ferdinand-lot/
Sleep Disorders Oxford Psychiatry Library Sue Wilson

https://ebookname.com/product/sleep-disorders-oxford-psychiatry-
library-sue-wilson/

Mandated Landscape British Imperial Rule in Palestine

1929 1948 1st Edition Roza El-Eini

https://ebookname.com/product/mandated-landscape-british-
imperial-rule-in-palestine-1929-1948-1st-edition-roza-el-eini/

Objects First with Java A Practical Introduction Using

BlueJ 1st edition Edition David J. Barnes

https://ebookname.com/product/objects-first-with-java-a-
practical-introduction-using-bluej-1st-edition-edition-david-j-
barnes/

Guiding the Behaviour of Children Young People 1st

Edition Jennie Lindon

https://ebookname.com/product/guiding-the-behaviour-of-children-
young-people-1st-edition-jennie-lindon/

The Presocratics Edward Hussey

https://ebookname.com/product/the-presocratics-edward-hussey/
The Tradition of the Syriac Church of Antioch
Concerning the Primacy and the Perogatives of St Peter
and of His Successors the Roman Pontiffs Syriac and
Latin Edition Syriac Studies Library Cyril Behnam Benni
https://ebookname.com/product/the-tradition-of-the-syriac-church-
of-antioch-concerning-the-primacy-and-the-perogatives-of-st-
peter-and-of-his-successors-the-roman-pontiffs-syriac-and-latin-
edition-syriac-studies-library-cyril-behn/
Multiresolution
Signal
Decomposition
Transforms, Subbands, and Wavelets

Second Edition
Series in Telecommunications
Series Editor
T, Russell Hsing
Bell Communications Research
Morristown, NJ

Multiresolution Signal Decomposition: Transforms, Subbands,

and Wavelets
Ali N. Akansu and Richard A. Haddad
New Jersey Institute of Technology
Newark, NJ

Other Books in the Series

Hseuh-Ming Hang and John W. Woods, Handbook of Visual Communications: 1995
John J. Metzner, Reliable Data Communications: 1997
Tsong-Ho Wu and Noriaki Yoshikai, ATM Transport and Network Integrity: 1997
Shuo-Yen and Robert Li, Algebraic Switching Theory and Broadband Applications: 1999
Winston I. Way, Broadband Hybrid Fiber Coax Access System Technologies: 1999
Multiresolution
Signal
Decomposition
Transforms, Subbands, and Wavelets

Second Edition

All N. Akansu
and

Richard A. Haddad
New Jersey Institute of Technology
Newark, NJ

ACADEMIC PRESS
A Horcourt Science and Technology Company

San Diego San Francisco New York Boston

London Sydney Tokyo
This book is printed on acid-free paper. (°°)

Copyright © 2001, 1992 by Academic Press

All rights reserved..

No part of this publication may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopy, recording, or any information storage and
retrieval system, without permission in writing from the publisher.

Requests for permission to make copies of any part of the work should be mailed to the
following address: Permissions Department, Harcourt, Inc., 6277 Sea Harbor Drive, Orlando,
Florida 32887-6777.

ACADEMIC PRESS
A Harcourt Science and Technology Company
525 B Street, Suite 1900, San Diego, CA 92101-4495 USA
http://www.academicpress.com

Academic Press
Harcourt Place, 32 Jamestown Road, London NW1 7BY UK

Library of Congress Catalog Number: 99-68565

International Standard Book Number: 0-12-047141-8

Printed in the United States of America

00 01 02 03 04 EB 9 8 7 6 5 4 3 2 1
To Bilge and Elizabeth
This page intentionally left blank
Contents

1 Introduction 1
1.1 Introduction 1
1.2 Why Signal Decomposition? 2
1.3 Decompositions: Transforms, Subbands, and Wavelets 3
1.3.1 Block Transforms and Filter Banks 4
1.3.2 Multiresolution Structures 7
1.3.3 The Synthesis/Analysis Structure 8
1.3.4 The Binomial-Hermite Sequences: A Unifying
Example 9
1.4 Performance Evaluation and Applications 9

2 Orthogonal Transforms 11
2.1 Signal Expansions in Orthogonal Functions 12
2.1.1 Signal Expansions 12
2.1.2 Least-Squares Interpretation 17
2.1.3 Block Transforms 19
2.1.4 The Two-Dimensional Transformation 24
2.1.5 Singular Value Decomposition 26
2.2 Transform Efficiency and Coding Performance 30
2.2.1 Decorrelation, Energy Compaction, and the KLT 30
2.2.2 Comparative Performance Measures . 37
2.3 Fixed Transforms . 41
2.3.1 Sinusoidal Transforms , 42
2.3.2 Discrete Polynomial Transforms 55
2.3.3 Rectangular Transforms 65
2.3.4 Block Transform Packets 70
2.4 Parametric Modeling of Signal Sources 71
2.4.1 Autoregressive Signal Source Models 72
vii
viii CONTENTS
2.4.2 AR(1) Source Model 73
2.4.3 Correlation Models for Images 74
2.4.4 Coefficient Variances in Orthogonal Transforms 76
2.4.5 Goodness of 2D Correlation Models for Images 80
2.4.6 Performance Comparison of Block Transforms 81
2.5 Lapped Orthogonal Transforms 86
2.5.1 Introduction 86
2.5.2 Properties of the LOT 88
2.5.3 An Optimized LOT 90
2.5.4 The Fast LOT 93
2.5.5 Energy Compaction Performance of the LOTs 95
2.6 2D Transform Implementation 97
2.6.1 Matrix Kronecker Product and Its Properties 97
2.6.2 Separability of 2D Transforms 99
2.6.3 Fast 2D Transforms 101
2.6.4 Transform Applications 102
2.7 Summary 103

3 Theory of Subband Decomposition 113

3.1 Multirate Signal Processing 114
3.1.1 Decimation and Interpolation 114
3.1.2 Polyphase Decomposition . 123
3.2 Bandpass and Modulated Signals 128
3.2.1 Integer-Band Sampling 129
3.2.2 Quadrature Modulation 129
3.3 Mth Band, Mirror, & Power Complementary Filters 134
3.3.1 Mth Band Filters 134
3.3.2 Mirror Image Filters 135
3.3.3 Power Complementary Filters 137
3.4 Two-Channel Filter Banks 137
3.4.1 Two-Channel PR-QMF Bank 138
3.4.2 Regular Binary Subband Tree Structure 141
3.4.3 Irregular Binary Subband Tree Structure 146
3.4.4 Dyadic or Octave Band Subband Tree Structure 148
3.4.5 Laplacian Pyramid for Signal Decomposition 149
3.4.6 Modified Laplacian Pyramid for Critical Sampling 152
3.4.7 Generalized Subband Tree Structure 155
3.5 M-Barid Filter Banks 156
3.5.1 The M-Band Filter Bank Structure 158
CONTENTS ix
3.5.2 The Polyphase Decomposition 161
3.5.3 PR Requirements for FIR Filter Banks 170
3.5.4 The Paraunitary FIR Filter Bank . 171
3.5.5 Time-Domain Representations 180
3.5.6 Modulated Filter Banks 190
3.6 Cascaded Lattice Structures 193
3.6.1 The Two-Band Lossless Lattice 194
3.6.2 The M-Band Paraunitary Lattice 197
3.6.3 The Two-Band Linear-Phase Lattice 199
3.6.4 M-Band PR Linear Phase Filter Bank 203
3.6.5 Lattice Realizations of Modulated Filter Bank 206
3.7 IIR Subband Filter Banks 211
3.7.1 All-Pass Filters and Mirror Image Polynomials . 213
3.7.2 The Two-Band IIR QMF Structure 216
3.7.3 Perfect Reconstruction IIR Subband Systems 218
3.8 Transmultiplexers 226
3.8.1 TDMA, FDMA, and CDMA Forms of the
Transmultiplexor 227
3.8.2 Analysis of the Transmultiplexor 231
3.8.3 Orthogonal Transmultiplexor 235
3.9 Two-Dimensional Subband Decomposition 236
3.9.1 2D Transforms and Notation 236
3.9.2 Periodic Sequences and the DFT 237
3.9.3 Two-Dimensional Decimation and Interpolation 240
3.9.4 The 2D Filter Bank 245
3.9.5 Two-Band Filter Bank with Hexagonal or Quincunx Sampling251
3.9.6 Fan Filter Banks 258
3.10 Summary . 259

4 Filter Bank Families: Design and Performance 271

4.1 Binomial QMF-Wavelet Filters . 271
4.1.1 Binomial QMF and Orthonormal Wavelets 276
4.2 Maximally Flat Filters 278
4.3 Bernstein QMF-Wavelet Filters 281
4.4 Johnston QMF Family 286
4.5 Smith-Barnwell PR-CQF Family . 286
4.6 LeGall-Tabatabai PR Filter Bank 289
4.7 Princen-Bradley QMF 292
4.8 Optimal PR-QMF Design for Subband Image Coding . 292
x CONTENTS

4.8.1 Parameters of Optimization . 293

4.8.2 Optimal PR-QMF Design: Energy Compaction 297
4.8.3 Optimal PR-QMF Design: Extended Set of Variables . . . 297
4.8.4 Samples of Optimal PR-QMFs and Performance 298
4.9 Performance of PR-QMF Families 304
4.10 Aliasing Energy in Multiresolution Decomposition 308
4.10.1 Aliasing Effects of Decimation/Interpolation 308
4.10.2 Nonaliasing Energy Ratio 313
4.11 GTC and NER Performance 314
4.12 Quantization Effects in Filter Banks 315
4.12.1 Equivalent Noise Model 316
4.12.2 Quantization Model for M-Band Codec 318
4.12.3 Optimal Design of Bit-Constrained, pdf-Optimized Filter
Banks 323
4.13 Summary 324

5 Time-Frequency Representations 331

5.1 Introduction 331
5.2 Analog Background—Time Frequency Resolution 332
5.3 The Short-Time Fourier Transform 341
5.3.1 The Continuous STFT 342
5.3.2 The Discrete STFT 343
5.3.3 ThexDiscrete-Time STFT, or DFT 345
5.4 Discrete-Time Uncertainty and Binomial Sequences 347
5.4.1 Discrete-Time Uncertainty 347
5.4.2 Gaussian and Binomial Distributions 350
5.4.3 Band-Pass Filters 353
5.5 Time-Frequency Localization 355
5.5.1 Localization in Traditional Block Transforms 355
5.5.2 Localization in Uniform M-Band Filter Banks 355
5.5.3 Localization in Dyadic and Irregular Trees 360
5.6 Block Transform Packets 362
5.6.1 From Tiling Pattern to Block Transform Packets 364
5.6.2 Signal Decomposition in Time-Frequency Plane 373
5.6.3 From Signal to Optimum Tiling Pattern . 376
5.6.4 Signal Compaction 380
5.6.5 Interference Excision 382
5.6.6 Summary 384
CONTENTS xi
6 Wavelet Transform 391
6.1 The Wavelet Transform 392
6.1.1 The Continuous Wavelet Transform 392
6.1.2 The Discrete Wavelet Transform 396
6.2 Multiresolution Signal Decomposition . 401
6.2.1 Multiresolution Analysis Spaces , 402
6.2.2 The Haar Wavelet 404
6.2.3 Two-Band Unitary PR-QMF and Wavelet Bases 411
6.2.4 Multiresolution Pyramid Decomposition , . 416
6.2.5 Finite Resolution Wavelet Decomposition . 421
6.2.6 The Shannon Wavelets . 422
6.2.7 Initialization and the Fast Wavelet Transform . 425
6.3 Wavelet Regularity and Wavelet Families 427
6.3.1 Regularity or Smoothness 427
6.3.2 The Daubechies Wavelets 430
6.3.3 The Coiflet Bases 431
6.4 Biorthogonal Wavelets and Filter Banks . 432
6.5 Discussions and Conclusion , , 437

7 Applications 443
7.1 Introduction 443
7.2 Analysis/Synthesis Configuration 444
7.2.1 Selection of Analysis and Synthesis Filters . 445
7.2.2 Spectral Effects of Down- and Up-samplers . 446
7.2.3 Tree Structuring Algorithms for Hierarchical Subbarid Trans-
forms 447
7.2.4 Subband Coding .448
7.2.5 Interference Excision in Direct Sequence Spread Spectrum
Communications 452
7.3 Synthesis/Analysis Configuration 457
7.3.1 Discrete Multitone Modulation for Digital
Communications 459
7.3.2 Spread Spectrum PR-QMF Codes for CDMA
Communications 463

A Resolution of the Identity and Inversion 473

xii CONTENTS
B Orthonormality in Frequency 477

C Problems 479
Preface

Since the first edition of this book in 1992 we have witnessed a flood of books-
texts, monographs, and edited volumes describing different aspects of block trans-
forms, multirate filter banks, arid wavelets. Some of these have been mathe-
matically precise, designed for the rigorous theoretician, while others sought to
interpret work in this arena for engineers and students.
The field is now mature, yet active. The theory is much better understood
in the signal processing community, and applications of the multiresolution con-
cept to situations in digital multimedia, communications, and others abound. In
the first edition and in the early days of multirate filter banks a prime empha-
sis was on signal compaction and coding. Today, multiresolution decomposition
and time-frequency concepts have opened up new vistas for further development
and application. These ideas concerning orthogonal signal analysis and synthesis
have led to applications in digital audio broadcasting, digital data hiding and wa-
termarking, wireless and wireline communications, audio and video coding, and
many others.
In this edition, we continue to treat block transforms, subband filter banks,
and wravelets from a common unifying standpoint. We demonstrate the com-
monality among these signal analysis and synthesis techniques by showing how
the block transform evolves gracefully into the more general multirate subband
filter bank, arid then by establishing the multiresolution decomposition features
common to both the dyadic subband tree structure and the orthonormal wave-
let transform. In order to achieve this unification, we have focused mainly on
orthonormal decompositions and presented a unified and integrate*! treatment of
multiresolution signal decomposition techniques using the property of orthonor-
rnality as the unifying theme.(A few exceptions, such as the oversampled Laplacian
pyramid and biorthogonal filter banks are also presented because they provide an
historical perspective and serve as foils to the critically sampled, orthonormal
subband structures we emphasize.)

xin
xiv Preface
Our second focus in the first edition was the application of decomposition
techniques to signal compression and coding. Accordingly, we describe objcctiw
performance criteria that measure this attribute and then compare the different
techniques on this basis. We acknowledge that subjective evaluations of decompo-
sition are important in applications such as image and video processing and coding,
machine vision, and pattern recognition. Such aspects are treated adequately in
the literature cited and are deemed beyond the scope of this book. A new fo-
cus in this edition is the time-frequency properties of signals and decomposition
techniques. Accordingly, this text provides tables listing the coefficients of pop-
ular block transforms, subband and wavelet filters, and also their time-frequency
features and compaction performance for both theoretical signal models and stan-
dard test images. In this respect, we have tried to make the book a reference text
as well as a didactic monograph.
Our approach is to build from the fundamentals, taking simple representative-
cases first and then extending these to the next level of generalization. For exam-
ple, we start with block transforms, extend these to lapped orthogonal transforms,
arid then show both to be special cases of subband filter structures. We have
avoided the theorem-proof approach, preferring to give explanation and deriva-
tions emphasizing clarity of concept rather than strict rigor.
Chapter 2 on orthogonal transforms introduces block transforms from a least-
squares expansion in orthogonal functions. Signal models and decorrelation and
compaction performance measures are then used to evaluate and compare several
proposed block and lapped transforms. The biorthogonal signal decomposition is
mentioned.
Chapter 3 presents the theory of perfect reconstruction, orthonormal two-band
and M-band filter banks with emphasis on the finite impulse response variety. A
key contribution here is the time-domain representation of an arbitrary multirate
filter bank, from which a variety of special cases emerge—paraunitary, biorthog-
onal, lattice, LOT, and modulated filter banks. The two-channel, dyadic tree
structure then provides a multiresolution link with both the historical Laplacian
pyramid and the orthonormal wavelets of Chapter 6. A new feature is the rep-
resentation of the transmultiplexer as the synthesis/analysis dual of the analy-
sis/synthesis multirate filter bank configuration.
Chapter 4 deals with specific filter banks and evaluates their objective per-
formance. This chapter relates the theory of signal decomposition techniques
presented in the text with the applications. It provides a unified performance
evaluation of block transforms, subband decomposition, and wavelet filters from
a signal processing and coding point of view. The topic of optimal filter banks
presented in this chapter deals with solutions based on practical considerations
Preface xv
in image coding. The chapter closes with the modeling and optimum design of
quantized filter banks.
Chapter 5 on time-frequency (T-F) focuses on joint time-frequency properties
of signals and the localization features of decomposition tools. There is a discussion
of techniques for synthesizing signals and block transforms with desirable T-F
properties and describes applications to compaction and interference excision in
spread spectrum communications.
Chapter 6 presents the basic theory of the orthonormal and biorthogonal wave-
let transforms and demonstrates their connection to the orthonormal dyadic sub-
band tree of Chapter 3. Again, our interest is in the linkage to the multiresolution
subbaiid tree structure, rather than with specific applications of wavelet trans-
forms.
Chapter 7 is a review of recent applications of these techniques to image coding,
and to communications applications such as discrete rnultitone (DMT) modula-
tion, and orthogonal spread spectrum user codes. This chapter links the riches
of linear orthogonal transform theory to the popular and emerging transform ap-
plications. It is expected that this linkage might spark ideas for new applications
that, benefit from these signal processing tools in the future.
This book is intended for graduate students and R&D practitioners who have
a working knowledge of linear system theory and Fourier analysis, some linear
algebra, random signals arid processes, and an introductory course in digital signal
processing. A set of problems is included for instructional purposes.
For classroom presentation, an instructor may present the material in the text
in three packets:
(1) Chapters 2 and 5 on block transforms and time-frequency methods
(2) Chapters 3 arid 4 on theory and design of rnultirate filter banks
(3) Chapters 6 and 7 on wavelets and transform applications
As expected, a book of this kind would be impossible without the cooperation
of colleagues in the field. The paper preprints, reports, and private communica-
tions they provided helped to improve significantly the quality and timeliness of
the book. We acknowledge the generous help of N. Sezgin and A. Bircari for some
figures. Dr. T. Russell Hsing of Bellcore was instrumental in introducing us to
Academic Press. It has been a pleasure to work with Dr. Zvi Ruder during this
project. Dr. Eric Viscito was very kind to review Chapter 3. The comments and
suggestions of our former and current graduate students helped to improve the
quality of this book. In particular, we enjoyed the stimulating discussions and in-
teractions with H. Caglar, A. Benyassine, M. Tazebay, X. Lin, N. Uzun. K. Park.
K. Kwak, and J,C. Horng. We thank them all. Lastly, we appreciate and thank
xvi Preface
our families for their understanding, support, and extraordinary patience during
the preparation of this book.

AM N. Akansu
Richard A. Haddad
April 2000
Chapter 1

Introduction

1.1 Introduction
In the first edition of this book, published in 1992, we stated our goals as three-
fold:
(1) To present orthonormal signal decomposition techniques—transforms, sub-
bands, and wavelets—from a unified framework and point of view.
(2) To develop the interrelationships among decomposition methods in both time
and frequency domains and to define common features.
(3) To evaluate and critique proposed decomposition strategies from a compres-
sion coding standpoint using measures appropriate to image processing.
The emphasis then was signal coding in an analysis/synthesis structure or codec.
As the field matured and new insights were gained, we expanded our vistas to
communications systems and other applications where objectives other than com-
pression are vital — as for example, interference excision in CDMA spread spec-
trum systems. We can also represent certain communications systems such as
TDMA, FDMA, and CDMA as synthesis/analysis structures, i.e., the conceptual
dual of the compression codec. This duality enables one to view all these systems
from one unified framework.
The Fourier transform and its extensions have historically been the prime vehi-
cle for signal analysis and representation. Since the early 1970s, block transforms
with real basis functions, particularly the discrete cosine transform (DCT), have
been studied extensively for transform coding applications. The availability of
simple fast transform algorithms and good signal coding performance made the
DCT the standard signal decomposition technique, particularly for image and
video. The international standard image-video coding algorithms, i.e., CCITT
2 CHAPTER 1. INTRODUCTION

H.261, JPEG, and MPEG, all employ DCT-based transform coding.

Since the recent research activities in signal decomposition are basically driven
by visual signal processing and coding applications, the properties of the human
visual system (HVS) are examined and incorporated in the signal decomposition
step. It has been reported that the HVS inherently performs multiresolution sig-
nal processing. This finding triggered significant interest in multiresolution signal
decomposition and its mathematical foundations in mult irate signal processing
theory. The multiresolution signal analysis concept also fits a wide spectrum
of visual signal processing and visual communications applications. Lower, i.e..
coarser, resolution versions of an image frame or video sequence are often suffi-
cient in many instances. Progressive improvement of the signal quality in visual
applications, from coarse to finer resolution, has many uses in compute1!' vision,
visual communications, and related fields.
The recognition that multiresolution signal decomposition is a by-product of
rnultirate subband filter banks generated significant interest in the design of better-
performing filter banks for visual signal processing applications.
The wavelet transform with a capability for variable time-frequency resolution
has been promoted as an elegant multiresolution signal processing tool. It was
shown that this decomposition technique is strongly linked to subband decompo-
sition. This linkage stimulated additional interest in subband filter banks, since
they serve as the only vehicle for fast orthonormal wavelet transform algorithms
and wavelet transform basis design.

1.2 Why Signal Decomposition?

The uneven distribution of signal energy in the frequency domain has made signal
decomposition an important practical problem. Rate-distortion theory shows that
the uneven spectral nature of real-world signals can provide the basis for source
compression techniques. The basic concept here is to divide the signal spectrum
into its subspectra or subbands, and then to treat those subspectra individually
for the purpose at hand. From a signal coding standpoint, it can be appreciated
that subspectra with more energy content deserve higher priority or weight for
further processing. For example, a slowly varying signal will have predominantly
low-frequency components. Therefore, the low-pass subbands contain most of its
total energy. If one discards the high-pass analysis subbands and reconstructs the
signal, it is expected that very little or negligible reconstruction error occurs after
this analysis-synthesis operation.
1.3. DECOMPOSITIONS: TRANSFORMS, SUBBANDS, AND WAVELETS 3
The decomposition of the signal spectrum into subbands provides the mathe-
matical basis for two important and desirable features in signal analysis and pro-
cessing. First, the monitoring of signal energy components within the subbands
or subspectra is possible. The subband signals can then be ranked and processed
independently. A common use of this feature is in the spectral shaping of quanti-
zation noise in signal coding applications. By bit allocation we can allow different
levels of quantization error in different subbands. Second, the subband decomposi-
tion of the signal spectrum leads naturally to multiresolution signal decomposition
via multirate signal processing in accordance with the Nyquist sampling theorem.
Apart from coding/compression considerations, signal decomposition into sub-
bands permits us to investigate the subbands for contraband signals, such as band-
limited or single tone interference. We have also learned to think more globally to
the point of signal decomposition in a composite time-frequency domain, rather
than in frequency subbands as such. This expansive way of thinking leads natu-
rally to the concept of wavelet packets (subband trees), and to the block transform
packets introduced in this text.

1.3 Decompositions: Transforms, Subbands,

and Wavelets
The signal decomposition (and reconstruction) techniques developed in this book
have three salient characteristics:
(1) Orthonorrnality. As we shall see, the block transforms will be square unitary
matrices, i.e.. the rows of the transformation matrix will be orthogonal to each
other; the subband filter banks will be paraunitary, a special kind of orthonormal-
ity, and the wavelets will be orthonormal.
(2) Perfect reconstruction (PR). This means that, in the absence of encoding,
quantization, and transmission errors, the reconstructed signal can be reassem-
bled perfectly at the receiver.
(3) Critical sampling. This implies that the signal is subsampled at a minimum
possible rate consistent with the applicable Nyquist theorem. From a practical
standpoint, this means that if the original signal has a data rate of fs samples or
pixels per second, the sum of the transmission rates out of all the subbands is also
/.-
The aforementioned are the prime ingredients of the decomposition techniques.
However, we also briefly present a few other decomposition methods for contrast
or historical perspective. The oversampled Laplacian pyramid, biorthogonal filter
banks, and non-PR filter banks are examples of these, which we introduce for
4 CHAPTER 1. INTRODUCTION
didactic value.

(b)

Figure 1.1: (a) Analysis-synthesis structure; (b) synthesis/analysis system.

As shown iri Fig. 1.1 (a), the input signal x is decomposed in the analysis
section, encoded, and transmitted. At the receiver or synthesis section, it is re-
constructed as x. In a perfect reconstruction system x = x within an allowable
delay. In a critically sampled system, the sum of the data rates of the decomposed
signal components equals that of the input signal.
In Fig. l.l(b), the dual operation is shown. Typically, the synthesis section
could be a TDMA or FDMA multiplexer wherein several signals are separated
in time (TDMA), frequency (FDMA), or in time-frequency (CDMA), and com-
bined into one signal for transmission. The received signal is then separated into
components in the analysis section.

1.3.1 Block Transforms and Filter Banks

In block transform notation, the analysis or decomposition operation suggested
in Fig. 1.1 is done with a blockwise treatment of the signal. The input signal
is first segmented into nonoverlapping blocks of samples. These signal blocks or
vectors are transformed into spectral coefficient vectors by the orthogonal matrix.
The spectral unevenness of the signal is manifested by unequal coefficient energies
by this technique and only transform coefficients with significant energies need
be considered for further processing. Block transforms, particularly the discrete
1.3. DECOMPOSITIONS: TRANSFORMS, SUBBANDS, AND WAVELETS 5
cosine transforms, have been used in image-video coding. Chapter 2 introduces
and discusses block transforms in detail and provides objective performance evalu-
ations of known block transforms. The Karhunen-Loeve transform, or KLT, is the
unique input-signal dependent optimal block transform. We derive its properties
and use it as a standard against which all other fixed transforms can be compared.
In block transforms, the duration or length of the basis functions is equal to the
size of the data block. This implies that the transform and inverse transform ma-
trices are square. This structure has the least possible freedom in tuning its basis
functions. It can meet only an orthonormality requirement and, for the optimal
KLT, generate uncorrelated spectral coefficients. Limited joint time-frequency
localization of basis functions is possible using the concept of block transform
packets (Chapter 5).
More freedom for tuning the basis functions is possible if we extend the du-
ration of these functions. Now this rectangular transform or decomposition has
overlapping basis functions. This overlapping eliminates the "blockiness" problem
inherent in block transforms. Doubling the length of the basis sequences gives the
lapped orthogonal transform, or LOT, as discussed in Section 2.5.
In general, if we allow arbitrary durations for the basis sequence filters, the
finite impulse response (FIR) filter bank or subband concept is reached. Therefore,
block transforms arid LOTs can be regarded as special filter banks. The rnultirate
signal processing theory and its use in perfect reconstruction analysis-synthesis
filter banks are discussed in depth in Chapter 3. This provides the common frame
through which block transforms, LOTs, and filter banks can be viewed.
Figure 1.2 shows a hierarchical conceptual framework for viewing these ideas.
At the lowest level, the block transform is a bank of M filters whose impulse
responses are of length L = M. At the next level, the LOT is a bank of M filters,
each with impulse responses (or basis sequences) of length L = 2M. At, the top of
the structure is the M-band multirate filter bank with impulse responses of any
length L > M. On top of that is the M-band rnultirate filter bank with impulse
responses of arbitrary length L > M. This subband structure is illustrated in
Fig. 1.3(a), where the signal is decomposed into M equal bands by the filter bank.
The filter bank often used here has frequency responses covering the Af-bands
from 0 to /s/2. When these frequency responses are translated versions of a low-
frequency prototype, the bank is called a modulated filter bank.
The Nyquist theorem in a multiband system can now be invoked to subsample
each band. The system is critically subsampled (or maximally decimated) when
the decimation factor D or subsampling parameter equals the number of subbands
M. When D < M, the system is oversampled.
6 CHAPTER 1. INTRODUCTION

Figure 1.2: An overview of M-band signal decomposition.

Figure 1.3: Multirate filter bank with equal bandwidths: (a) M-band; (b) four-
band, realized by a two-level binary (regular) tree.
1.3. DECOMPOSITIONS: TRANSFORMS, SUBBANDS, AND WAVELETS 7
Another way of realizing the decomposition into M equal subbands is shown by
the hierarchical two-band subband tree shown in Fig. 1.3(b). Each level of the tree
splits the preceding subband into two equal parts, permitting a decomposition into
M — 2fc equal subbands. In this case the M-band structure is said to be realized
by a dilation of the impulse responses of the basic two-band structure at each level
of the tree, since splitting each subband in two dilates the impulse response by
this factor.

1.3.2 Multiresolution Structures

Yet another possible decomposition is shown in Fig. 1.4, which represents a ''dyadic
tree" decomposition. The signal is first split into low- and high-frequency com-
ponents in the first level. This first low-frequency subband, containing most of
the energy, is subsarnpled and again decomposed into low- and high-frequency
subbands. This process can be continued into K levels. The coarsest signal is
the one labeled LLL in the figure. Moving from right to left in this diagram, we
see a progression from coarser to finer signal representation as the high-frequency
"detail" is added at each level. The signal can thus be approximately represented
by different resolutions at each level of the tree.

Figure 1.4: Multiresolution dyadic tree: L and H represent low-pass and high-pass
filters, respectively.

An oversampled version of this tree, called the Laplacian pyramid, was first in-
troduced for image coding by Burt and Adelson (1983). These topics are explained
in detail and the reference is given in Chapter 3.
Wavelet transforms recently have been proposed as a new multiresolution de-
composition tool for continuous-time signals. The kernel of the wavelet transform
is obtained by dilation and translation of a prototype bandpass function. The
8 CHAPTER 1. INTRODUCTION
discrete wavelet transform (DWT) employs discretized dilation and translation
parameters. Simply stated, the wavelet transform permits a decomposition of a
signal into the sum of a lower resolution (or coarser) signal plus a detail, much
like the dyadic subband tree in the discrete-time case. Each coarse approximation
in turn can be decomposed further into yet a coarser signal and a detail signal at
that resolution. Eventually, the signal can be represented by a low-pass or coarse
signal at a certain scale (corresponding to the level of the tree), plus a sum of
detail signals at different resolutions. In fact, the subband dyadic tree structure
conceptualizes the wavelet multiresolution decomposition of a signal. We show in
Chapter 6 that the base or prototype function of the orthonormal wavelet trans-
form is simply related to the two-band unitary perfect reconstruction quadrature
mirror filters (PR-QMF), and that the fast wavelet transform algorithm can also
be strongly linked to the dyadic tree filter bank. Hence, from our perspective, we
view wavelets and dyadic subband trees as multiresolution decomposition tech-
niques of the continuous-time and discrete-time signals, respectively, as suggested
in Fig. 1.2.

1.3.3 The Synthesis/Analysis Structure

figure 1.5 shows the dual synthesis/analysis system. As mentioned earlier and ex-
plained in Chapter 3, this structure could represent any one of several multiplexing
systems depending on the choice of the synthesis and analysis filters. Interesting
enough, the conditions for alias cancellation in Fig. 1.3 and for zero cross-talk in
Fig. 1.5 are the same. Additionally, the conditions for "perfect reconstruction,"
x(n] = x(n — no) in Fig. l.l(a) and Xi(ri) = Xi(n — no) in Fig. l.l(b), are the same!

Figure 1.5: Transmultiplexer as a synthesis/analysis structure

1.4. PERFORMANCE EVAL UATION AND APPLICATIONS 9
1.3.4 The Rinomial-Hermite Sequences: A Unifying
Example
This book tries to provide a common framework for the interpretation arid evalu-
ation of all orthonormal signal decomposition tools: block transforms, subbands,
and wavelets. The Binomlal-Hermite sequences provide a family of functions with
applications that touch all these categories. This elegant family will be used as a
vehicle to illustrate and link together all these topics. At the simplest level they
provide a set of functions for orthogonal signal expansions. Suitably modified,
they generate block transform, called the modified Hermite transform (MHT).
Then we linearly combine members of the Binomial family to obtain the unique,
maximally fiat squared magnitude, two-band paraunitary QMF. These in turn
are recognized as the orthonormal wavelet filters devised by Daubechies. These
functions also play a major role as kernels in discrete time-frequency analysis.

1.4 Performance Evaluation and Applications

One of the objectives of this book is a comparative evaluation of several of the
more popular decomposition techniques. In Chapter 2, families of transforms
are described and their compaction properties are evaluated both from a block
transform and LOT realization. Chapter 4 presents the comparative evaluation
of known filter families and wavelet filters. It provides criteria by which all these
strategies can be compared.
Chapter 5 compares the time-frequency localization properties of block trans-
forms and wavelet filters.
In addition to comparative evaluations, in Chapter 4 we also introduce an
optimal design approach for filters wherein the design parameters are made part of
the performance criteria that can be optimized. These optimal solutions set upper
performance bounds for FIR subband decomposition in a manner conceptually
similar to the performance bound that the optimal transform, the KLT, sets for
all block transforms.
Chapter 7 describes a medley of applications of these techniques to the solu-
tions of problems in communications and multimedia.
This page intentionally left blank
Chapter 2

Orthogonal Transforms

The purpose of transform coding is to decompose a batch of correlated signal sam-

ples into a set of uncorrelated spectral coefficients, with energy concentrated in as
few coefficients as possible. This compaction of energy permits a prioritization of
the spectral coefficients, with the more energetic ones receiving a greater allocation
of encoding bits. For the same distortion level, the total number of bits needed to
transmit encoded spectral coefficients is less than the number needed to transfer
the signal samples directly. This reduction in bit rate is termed compression.
In this chapter we introduce the subject of orthogonal transforms—from the
standpoint of function expansions in orthogonal series. For the most part, the
signals and basis functions have finite support—that is, are of finite duration—
as required for block transforms. For completeness of presentation, however, we
also include signals of semi-infinite extent. We group the block transforms consid-
ered here in three broad categories—sinusoidal, polynomial, and rectangular and
describe leading members in each category.
The decorrelation and energy compaction properties of these transforms con-
stitute the central issues in the applicability of these transforms for signal coding
purposes. Of equal importance is the computational complexity associated with
the respective transformations. In the sequel, we formulate criteria that permit
comparative evaluation of the performance of different transforms in signal coding
applications. These measures are also used on standard test images.
At low bit rates, transform-coded images often exhibit a "blockiness" at the
borders. The lapped orthogonal transform (LOT) was introduced to counteract
this effect. The spectral coefficients are calculated using data windows that over-
lap batch boundaries. The LOTs are described and evaluated using the criteria
developed for transform coding. The block transform and LOT are interpreted as
special cases of the multirate filter bank introduced in Chapter 3.
11
12 CHAPTER 2. ORTHOGONAL TRANSFORMS
The signals arid theory developed here for the most part represent functions
of one variable — i.e., one dimensional (ID) signals. The extension to the mul-
tidimensional ease is usually straightforward for separable transforms. In feet,
in some instances, the Kronecker product expansion of a two dimensional (2D)
transformation permits a factorization and grouping of terms that simply is not
possible in ID.

2.1 Signal Expansions in Orthogonal Functions

The orthogonal expansion of a continuous variable function is a subject extensively
addressed in the classical literature (Sansone, 1959). Our focus is the expansion
of sampled signals, i.e., sequences {/(&)}. Milne (1949) briefly treats this subject.
During the decade of the 1970s, the study of the subject gained in intensity as
several authors Campanella arid Robinson (1977); Ahmed and Rao (1975) devel-
oped expansions for discrete-variable functions to meet the needs of transform
coding. Orthogonal expansions provide the theoretical underpinnings for these
applications.

2.1.1 Signal Expansions

Our task here is the representation of a sequence (or discrete-time or discrete-
variable signal) {/(&)} as a weighted sum of component sequences. These possess
special properties that highlight certain features of the signal. The most familiar
is

where the component sequence 6(n — k] is the Kronecker delta sequence:

Here the weights are just the sample values themselves—not very interesting.
Instead we seek to represent {/(&)} as a superposition of component sequences
which can extract identifying features of the signal in a compact way. But in any
event, the component sequences should be members of an orthogonal family of
functions.
In what follows, we borrow some geometric concepts from the theory of linear
vector spaces. To fix ideas, consider a sequence {/(&)} defined on the interval
2.1. SIGNAL EXPANSIONS IN ORTHOGONAL FUNCTIONS 13

0 < k < N — 1. We can think of {/(&)} as an N dimensional vector / and

represent it by the superposition1

The form of Eq. (2.3) is a finite-dimensional version of Eq. (2.1). The norm of
/ is denned as

The sequence {/(&)} is now represented as a point in the N dimensional Eu-

clidean space spanned by the basis vectors {e 0 ,e l 5 ..., e^v-i}- These basis vectors
are linearly independent, since the linear combination

can vanish only if CQ = c\ = • • • — C]v-i = 0 . Another way of expressing this is

that no one basis vector can be represented as a linear combination of the others.
We say that the two sequences {g(k}} and {/(&)} with the same support.
i.e., the interval outside of which the sequence is zero, are orthogonal if the inner
product vanishes, or

Clearly the finite dimensional Kronecker delta sequences are orthogonal, since

For notational convenience, we will use /„ and f ( n ) interchangeably.

14 CHAPTER 2. ORTHOGONAL TRANSFORMS
Moreover, the norm of each basis vector is unity,

Again, we note that the weights in the expansion of Eq. (2.3) are just the
sample values that, by themselves, convey little insight into the properties of
the signal. Suppose that a set of basis vectors can be found such that the data
vector / can be represented closely by just a few members of the set. In that
case, each basis vector identifies a particular feature of the data vector, and the
weights associated with the basis vectors characterize the features of the signal.
The simplest example of this is the Fourier trigonometric expansion wherein the
coefficient value at each harmonic frequency is a measure of the signal strength
at that frequency. We now turn to a consideration of a broad class of orthogonal
expansions with the expectation that each class can characterize certain features
of the signal.
Suppose we can find { x n ( k ) , 0 < n,k < N — 1}, a family of N linearly
independent sequences on the interval [0, Ar — 1]. This family is orthogonal if

where cn is the norm of {xn(&)}. (The asterisk denotes the complex conjugate.)
The orthonormal family is obtained by the normalization,

which in turn shows that

Any nontrivial set of functions satisfying Eq. (2.11) constitutes an orthonormal

basis for the linear vector space. Hence {/(&)} can be uniquely represented as

where
2.1. SIGNAL EXPANSIONS IN ORTHOGONAL FUNCTIONS

The proof of Eq. (2.13) is established by multiplying both sides of Eq. (2,12)
by (f)s(k) and summing over the k index. Interchanging the order of summation
and invoking orthonormality shows that

The set of coefficients {$s, 0 < s < N — 1} constitute the spectral coefficients
of {/(&)} relative to the given orthonormal family of basis functions. Classically,
these are called generalized Fourier coefficients even when {(j)n(k)} are not sinu-
soidal.
The energy in a signal sequence is denned to be the square of the norm. The
Parseval theorem, which asserts that
N-l N-l
(2.14)

can be proved by multiplying both sides of Eq. (2.12) by their conjugates and
summing over k. This theorem asserts that the signal energy is preserved under
an orthonormal transformation and can be measured by the square of the norm
of either the signal samples or the spectral coefficients.
As we shall see, one of the prime objectives of transform coding is the redis-
tribution of energy into a few spectral coefficients.
On a finite interval, the norm is finite if all samples are bounded on that
interval. On the other hand, convergence of the norm for signals defined on [0, oo)
or (—00,00) requires much more stringent conditions, an obvious necessary one
being f(k)\ —> 0, as k —> ±00. Sequences with finite energy are said to be L2.
The Z-transform provides an alternative signal description, which is partic-
ularly useful in a filtering context. To this end, let $ n (^) be the Z-transform
(one-sided or two-sided, as required by the region of support) of {(f)n(k)}. Now
the orthogonality relationship of Eq. (2.11) can be restated as the contour integral

n(*)*;(4)— ^ <*™-a, (2.15)

where the contour is taken on the unit circle of the .Z-plane. From the Cauchy
residue theorem, the sum of the residues in all the poles of the integrand within
16 CHAPTER 2. ORTHOGONAL TRANSFORMS
the unit circle must vanish for n ^ s. But on the unit circle, z = e3UJ. With this
substitution Eq. (2.15) becomes

The latter form permits us to generate families of orthogonal sequences by speci-

fying the behavior of <i>n(eja;). (See Problem 2.4.)

Figure 2.1: Orthonormal spectral analyzer as a multirate filter bank.

The form of Eq. (2.13) suggests that the spectral coefficients {On} for the
sequence {/(&)} can be measured by the spectral analyzer shown as Fig. 2.1. The
signal f(k) is fed into a bank of FIR filters whose impulse responses are time
reversed and translated basis sequences {4>*(N — 1 — k) = hr(k)}. The output of
the rth filter is the convolution

Sampling this output at n = N — I gives the coefficient Or — yr(N — 1). The

collection of sampled outputs at this time gives the spectral coefficient vector
0T — [#o, ••-, ^-i] f°r the first block of data f7 = [/o,..., /N-I]-
The circle with the downward-pointing arrow in this diagram indicates that the
output sequence of each filter is subsampled, i.e., every JVth sample is retained. If
2,1. SIGNAL EXPANSIONS IN ORTHOGONAL FUNCTIONS 17

the input {/(&)} is a continuing stream of data, the subsampled outputs at times
N — 1. 2N — 1, ... represent successive spectral coefficient vectors corresponding to
successive blocks of data. In Chapter 3, we will interpret Fig. 2.1 as a multirate
filter bank that functions as the front end of a subband coder.

2.1.2 Least- Squares Interpretation

The set of coefficients {On} in Eq. (2.13) also provides the least-squares ap-
proximation to {/(&)}. Suppose we want to approximate {/(&)} by a super-
position of the first L of the N basis sequences, using weighting coefficients
{7,;, '/' — 0, 1, . . . , L — 1}. Then the best least-squares choice for these coefficients
is

The proof is as follows. Let the approximation be

and the error is then

The {'jr} are to be chosen to minimize the sum squared error

Expanding the latter and invoking orthonormality2 gives

Next, setting the partial of JL with respect to 7S to zero gives

with solution

2
For convenience, we pretend that /(/c),7i, and 4>r(k) are real.
18 CHAPTER 2. ORTHOGONAL TRANSFORMS
When L = N, we note that JN — 0. Thus, the sum squared error in using
f(k] as an approximation to f(k) is minimized by selecting the weights to be the
orthonorrnal spectral coefficients.
This choice of coefBcients has the property of finality. This means that if we
wish to reduce the error by the addition of more terms in Eq. (2.18), we need not
recalculate the previously determined values of {7r}- Also, we can show

The resulting minimized error sequence is

* — \,s i — i_j

Thus, the error sequence {t(k}} lies in the space spanned by the remaining
basis functions

whereas the estimate {/(&)} lies hi Vi, the space spanned by {(j)s(k)J 0< s<
L — 1). The term V% is the orthogonal complement of Vi with the property that
every vector in 1/2 is orthogonal to every vector in Vi. Furthermore, the space
V of all basis vectors is just the direct sum3 of Vi and V%. It is easy to see that
{f-(k}} is orthogonal to {/(&)}, i.e.,

In fact, it can be shown that the orthogonality of error and approximant is neces-
sary and sufficient to minimize the sum squared error. (Prob. 2.1)
A simple sketch depicting this relationship is shown in Fig. 2.2 for the case
JV — 3,L = 2. This sketch demonstrates that {/(&)}, the least squares approxi-
mation to {/(&)} is the orthogonal projection of {/(A:)} onto the two-dimensional
subspace spanned by basis sequences {0i(fc)|, (02 (A')}- For complex valued se-
quences {/*(&)} is used in Eq. (2.24).
Before closing this section, we note that all of these results and theorems are
valid for infinite dimensional spaces as well as finite dimensional ones, as long
as the norms of the sequences are bounded, i.e., are L2. We also note for later
reference two additional theorems to be used subsequently. First, the Cauchy-
Schwarz inequality asserts that
3
In our representation V = R ,the set of all real N tuples. The direct sum is V — Vi © V% if
and only if Vi f~l V2 = <£, and V = Vi U V2.
 Another Random Document on
Scribd Without Any Related Topics
 "Oh, cousin Bess, we've got lots of specimens!" shouted Rob from afar,
and Ted added,—

"A good start for our museum, sure enough!" while the boys settled
themselves on the piazza rail, and pulled various boxes from their pockets.

"Here are five moth-millers and a hornbug," announced Sam, producing

some rather dilapidated specimens, for the hornbug had lost three legs, and
of the moths, one was minus both antennæ, and another had a great slit in
one wing.

"I've got eight moths," said Phil. "I picked them up under the electric
light. They're real good ones, only they are singed a little. I s'pose that's
what killed them. And here," diving into his trousers pocket, "is a
bumblebee my father killed yesterday. Oh, dear! His head's come off. Can't
we stick it on?"

Ted had also brought his share of mangled veterans, and Rob showed
three or four moths, quite well prepared, and a pair of golden yellow June
bugs.

It was with some difficulty that Bessie preserved her gravity as she saw
the ruins spread out before her. But, always mindful that much of her
influence over her boys lay in her hearty sympathy in all their hobbies, she
looked them over with an air of deep interest, and then sent Rob into the
library for a certain book that not only had fine pictures of all sorts and
conditions of insects, but also gave full instructions for their capture and
preservation.

"If you are going to do anything about it, my boys," said she, "you
would much better start in the right way at the very first, so we can have a
really good collection; and then we can try to have a full one, of all the
insects in the region. If you must collect, this is better than the barbarous,
cruel habit so many boys have of stealing birds' eggs, too often nest and all.
The eggs themselves won't teach you anything about the birds; while from
these, you can get some idea of the life and habits of these little creatures."
 "Just look at this one!" exclaimed Ted, pouncing on one of the two
perfect specimens, a pale green lunar moth. "Oh, dear! There goes one of
his wings. What are these fellows so brittle for?"

"That is another thing. You mustn't handle these specimens or they will
break. Now, let us see what we can find out about them."

And the next hour was spent in a pleasant talk about the form and habits
of these tiny creatures, a talk that the boys never forgot, for it taught them,
for the first time, the great plan of creation, that develops in each living
creature the bodily form which will help it most to live its little life, an
explanation so clearly and vividly given that even Fred felt no need of the
pictures to understand the mechanism of their small bodies.

The collecting fever spread, and the boys were often seen skipping
about the fields, or plunging headlong over fences, net in hand, in pursuit of
some gaudy butterfly. Bessie tried faithfully to make the boys feel that the
main object was not the catching and killing the insects; but that this was
only to help them to a fuller understanding of the nature and varieties of
their prey. Their whole energy was directed in the line of insects, and boxes
of specimens so rapidly collected, that the prospect was that the whole
Carter family would soon have to move out of the house, to make room for
the army of moths and beetles, cocoons and butterflies, that speedily
accumulated. Even long-suffering Mrs. Carter protested when, one day, on
the piazza, she chanced to knock down a box containing a huge green worm
that Rob had carefully provided with food and air-holes, and shut up, in the
hope that he would spin himself into a chrysalis; for, as the cover of the box
fell off, out dropped, not only the captive worm, now dead, but also a
multitude of little yellow wrigglers, that quickly squirmed away.

"The worst of it all is," said Bess ruefully, when her mother brought up
the subject, "people seem to think that I am having this done for my own
especial pleasure and profit. I don't see what they think I want of them,
unless I collect them, as the Chinese do the bones of their ancestors and
friends, to bury them in some particular, consecrated spot. I was writing the
other day, in a great hurry to finish my letter to Mr. Allen in time for the
steamer, when Bridget came up to my room, and said some little girls
wanted to see me. I went down to the back door, and there stood the five
Tracy children, in a row. As soon as I appeared, the oldest, who acted as
spokeswoman, came forward and solemnly presented me with three tattered
butterflies. I had such hard work to be just grateful enough to satisfy them,
and yet not encourage them to bring me any more. And the last time Mrs.
Walsh called, that day you were out, she produced a small box that held a
common little white moth, and told me Bert said that I wanted all those I
could get, so she had brought me that one. Well, laddie, what now?" she
added, as Fred came into Mrs. Carter's room, where they were sitting.

"There's a boy down-stairs," he replied, "that wants to see you. I don't

know who he is. He saw me on the piazza, and went round there to me."

"I wonder who he is," said Bess, as she laid down her work and went
out of the room.

She soon came up again, looking both amused and disgusted.

"Another!" she exclaimed, as she took up her sewing.

"What is it now?" asked her mother, laughing.

"It was that little red-haired Irish boy that lives in behind the church. I
don't know what his name is, but alas! he knows me. He came to bring me
some twigs, apple, I should think, and on each one was a horrid great
worm"—and Bess shivered at the recollection—"covered with red and
yellow bristles, I told him I was much obliged, but I really didn't know
where I could keep them; but the poor little fellow—I shouldn't think he
was more than seven or eight—looked so disappointed that I finally took
them. I grieve to say that I cremated them as soon as he had gone. Fred, if
you love me, do, oh, do tell the boys that we only want dead specimens. My
plan was for a museum, not a menagerie; but if matters go on like this, the
club will soon become my bugaboo."
 CHAPTER XII.

THEIR SUMMER OUTING.

"ISLAND DEN," THOUSAND ISLANDS,

July 27, 18—.

My Dear Bess,—I know you always have been a good, kind-hearted

little soul, and now I am going to throw myself on your benevolence and
ask a favor of you. Say yes, that's a dear little sister! It is just this that I
want,—a two weeks' visit from you. "Island Den" was never half so cosy as
this summer, and there were never half so many pleasant people over at the
hotel. The change will do you good, and I have already heard from mother,
saying that she can spare you as well as not. Jack and the children want to
see you as badly as I do.

But as long as I know you'll never consent to drop all care—you've had
too much these last months for a young thing like you—and leave that boy
of yours at home, as would be ever so much better for you, bring him with
you, if you think he will be contented here. Jack says two boys take up no
more room than one, and Rob had better come too, to be company for him
after we have talked each other to death. Isn't he impertinent? But it is a
good idea, for they will amuse each other and leave us more time. Rob has
never been here, and I am quite curious to see your other charge. Do hurry
to come, for I am impatient to see you. I should think you might start by the
first of next week.

Jack wishes me to enclose these tickets for the journey, as a last

inducement. He says I am to tell you that they will be wasted unless you use
them, and that will be sure to bring you, as your frugal soul cannot bear to
waste anything.

I won't say any more, for you will be here so soon; and then how we
will talk!

Your loving sister,

ALICE.
 This was the letter which had caused a sensation in the Carter
household. Alice Carter, ten years older than Bess, had married a wealthy
New York banker, and was now the mother of two little girls. "Island Den,"
their luxurious summer home, was on one of the Thousand Islands, whither
for years they had gone to spend the months of July and August, and keep
open house for their friends.

It was now three years since Bess had been able to accept her annual
invitation to go there, for it was an expensive little trip, and of late some
treacherous Western loans had decidedly lessened her father's income, and
reduced the family from the comfortable position of doing just about as
their rather simple inclinations led them, to the need of carefully counting
the smaller expenses that so quickly absorb money,—no marked change,
only they did not travel quite as much, nor keep a horse and carriage, nor
have quite so many gowns, while those they had they made themselves. The
more than liberal sum that Mr. Allen was paying them for the board and
care of Fred was far more helpful than he had realized when he had made
them the offer, although the money bargain had been by no means a
determining cause in their taking Fred into their home. And, this year, Bess
had felt that it would be more than ever impossible for her to go away to
leave Fred, both on her mother's account and the boy's own, for the child
clung to her more and more closely, with a devotion touching to see.

But Alice and Jack had smoothed away every difficulty, and Bess, with
her conscience at rest, could now accept their threefold invitation. Now
there was a prospect of change, the girl admitted to herself that she was a
little tired, and well she might be, for, in addition to her other duties, she
had given constant thought and care, as well as much time and countless
steps, to the boy who had so grown to depend upon her. But if, at the close
of a long day, the thought of her own weariness ever crossed her mind, the
memory of all that the child had lost, and of the brave fight he was making
against the burden of his blindness, made her scorn the thought of self, as
unworthy of the courage and patient endurance she was daily preaching to
the child, and gave her new strength to go on.
 Rob was in raptures over the prospective journey, and, during the week
before they were to start, he made almost hourly calls on Bess, to see how
her preparations were coming on. The morning after he was told of his
invitation and its acceptance, he was up early, and, before breakfast, had
gone into the attic, scattered over the floor the usual contents of a small
trunk, long past its days of active service and now only used for storage,
and secretly conveyed the trunk to his own room. By dinner-time, many of
his possessions were stowed away in its depths; books, games, his air-rifle,
several yards of mosquito netting for butterfly-nets, a choice collection of
fish-hooks, and an odd assortment of strings and small articles of hardware
that filled it to the brim, leaving room for not so much as a single
handkerchief. Each day he added to his hoard, to the amusement of his
mother, who let him have his way until the final packing, when she should
bring order out of chaos.

Fred scarcely looked forward to their going with as much pleasure as

Rob, for at the idea of the journey and of meeting so many strangers, his
shy sensitiveness returned in all its force, and he would gladly have spent
the time alone with the servants at his father's house, rather than run the
gauntlet of the curious and thoughtless, though not unkind comments that
always met him when he went among strangers.

However, it was a merry party that, one cloudy August morning, Mr.
Carter escorted as far as Boston, and settled in the train for Albany, where
they were to change to a sleeper. Rob, in a light summer suit, armed with a
jointed fishing-pole and his tennis racket, his mother's compromise in the
affair of the trunk, led the way into the car. Mr. Carter followed with a lunch
basket of noble proportions, for experience had taught Bessie that boy
appetites are unfailing, and, on Fred's account, she dared not depend on
railway dining-rooms. Bess, with Fred, brought up the rear of the
procession. Rob was bubbling over with fun and nonsense, so that Fred
caught his spirit and answered jest with jest. As Mr. Carter left them, Bess
turned and surveyed her charges with a feeling of almost maternal pride.
Two more bonnie boys it would have been hard to find that day.

"I wonder if I look like their mother, or what people think I am," she
thought, as she looked from the quiet boy at her side to the lively one
opposite her. "I don't care very much— Oh, Rob, be careful," she exclaimed
aloud, as that youth, in changing the position of his fishing-pole, recklessly
battered the rear of the respectable black bonnet worn by an old woman in
front of him.

Rob instantly turned to offer a meek apology, but it had no effect on the
irate woman, who grasped her bonnet firmly with both hands, as she
exclaimed,—

"Needn't knock a body's head off! Folks shouldn't take boys on the
keers till they know how to behave!"

"I am very sorry, ma'am," ventured Rob again.

"So you'd ought to be!" was the snappish rejoinder. "I hope you are
ashamed of yourself to go hitting a woman old enough to be your mother
with your nuisancing contraptions!" Then, with a backward glance, she
added, as if to herself, "That other one looks more as if he'd behave himself
somehow. I guess I'll move round and set behind him."

And she gathered up her belongings and moved back, where the worthy
soul lent an attentive ear to all their conversation, and watched Fred with
curious eyes, while from time to time she scowled disapprovingly on Rob,
who was quite subdued by his misadventure.

Of course, Rob wished to take a lunch before they were fairly outside of
Boston, and, equally of course, he desired to patronize every trip of the
newsboy, and the vender of prize packages of cough candy, each one of
which was warranted to contain a rich jewel; but on these small points Bess
was firm, and he abandoned himself to the alternate pleasures of gazing out
at the car window at the miles of back doors, each filled with a family as
much interested in the train as if it were some rare and curious object, and
of inspecting his fellow-passengers, the usual assortment. Across from them
was a young Japanese, who had intensified the effect of his swarthy skin by
mounting a white felt hat. With him sat a man who was so drowsy that his
head constantly dropped forward on the round silver knob that headed his
cane, at the imminent risk of putting out his eyes. The force of the blow
never failed to waken him, and he straightened himself up with a sheepishly
defiant air, as if to refute any possible denial of his wakefulness. Behind
him sat a spinster of sixty, with lank side curls and a fidgety manner of
moving her satchel about. There was the usual number of commercial
travellers—why have they appropriated the name?—who, with their silk
hats carefully put away in the racks, and replaced by undignified skull-caps,
took out their note-books and wrote up the record of their last sales; there
was the usual Irish mamma with five small children, who walked the entire
length of the car and planted herself in the little corner seat next the door,
with her offspring about her, budget in hand, ready to leave the train at a
moment's notice; and there were a few young women, each absorbed in her
novel or magazine, whom Rob surveyed with disfavor, as not being as
pretty as cousin Bess.

Leaning far forward, he was just describing some of these people for
Fred's benefit, when a sudden voice behind them made all three of the party
start. It was the woman whose bonnet Rob had hit.

"I want to know what's the matter with that 'ere boy," she demanded in
no gentle tone, as she pointed at Fred. "Can't he see, or what on airth's the
matter with him?"

Poor Fred! His laugh died away, and, turning very white, he leaned back
in his corner, while Bess answered their inquisitive neighbor with an icy
politeness, as she gave the boy's hand an encouraging pat. The brutal
abruptness of the question was more than the child could bear, and it was
long before he could speak or join in the conversation. Rob, meanwhile,
was vowing vengeance. His opportunity soon came.

Directly in front of him, in the seat vacated by his enemy, sat a middle-
aged man, who was carrying in his pocket a small gray kitten, probably a
gift to some child at home. Rob had noticed the little animal as the
gentleman came in, and from time to time he had turned to peep over at it,
when its owner was absorbed in his reading. At length the man laid aside
his paper, and turned to give his attention to the cat, which, however, was
nowhere to be found. He began to search about for it, looking rather
anxious. A sudden, naughty idea flashed into Rob's brain. Rising with an air
of polite sympathy, he inquired in a loud and cheerful voice,—
 "Can't I help you, sir? Which was it, a rattler, or just a common snake?"

The effect was instantaneous.

"Massy on us!" piped the aged heroine of the bonnet. "Snakes! Ow!"
And she climbed nimbly up on the seat, an example quickly followed by
her opposite neighbor. And though the cat was soon found and exhibited,
the two worthy women sat sideways on the seat, their feet and skirts
carefully tucked up beside them, until they left the train at Albany.

"Rob, how could you?" said Bess reprovingly, when quiet was restored.

"I don't care, cousin Bess. She was so mean to Fred that I did it on
purpose, and I sha'n't say I am sorry."

And Bess prudently changed the subject.

After a long delay at Albany, our travellers settled themselves anew in

their sleeper. Neither of the boys had ever before travelled all night, and it
seemed so cosy to go gliding away through the darkness that was slowly
shutting in the landscape. There were few people in the car, and Rob
prowled up and down, investigating his quarters, and making the
acquaintance of the porter; while Bess chatted with Fred, at ease once more
now that his dreaded neighbor had departed.

"I wish people wouldn't say such things," he told Bess. "Once in a while
I forget, but somebody always reminds me again, and it just makes me feel
as if everybody was watching me."

"It was a cruel question, cruelly asked," said Bess with some energy, as
she pulled off her gloves and took off her hat, preparatory to a comfortable
evening. "If people only knew how such remarks hurt! I wish I could save
you from them, laddie."

At this moment, Rob came back to his seat, and remarked with
conscious, but impenitent pride,—

"Didn't I just pay up that old woman? Mean old thing!"

 Then he devoted his attention to the porter, as he converted the seats
into diminutive bedrooms, partitioned and curtained off and sumptuously
furnished with a mirror and a wall pocket.

Long after the boys were stowed away for the night, Bess could hear
them whisper and giggle when a particularly loud snore from their next
neighbor broke the stillness; and at each stopping-place she heard Rob's
curtain fly up, to let him look out on the silent towns.

"Doesn't our Bess look matronly!" exclaimed Alice Rogers the next
morning, when she saw Bess and her two companions coming towards her.
"That one with her must be Fred Allen. Isn't he stunningly handsome,
Jack?"

"Poor little cub!" said Jack sympathetically, as he hurried forward to

meet them.

After the first confused moment of greeting and hand-shaking, question

and answer, Alice, a plump blonde who still kept much of her girlish beauty,
turned to the boys.

"Can this be my little cousin Rob, grown up to this?" she said, as she
kissed him, to his secret disgust, for Rob scorned kisses except from Bess.
"And this, I think, is Bessie's adopted boy, Fred, isn't it? I am so glad to
have you both here, for I like boys almost as well as Bess does."

Two days later, Rob sat on the piazza at Island Den, painfully fulfilling
his promise to write to his mother. Near him, Fred was swinging in a
hammock, holding beside him the two-year-old daughter of the house. Little
Alice had taken a violent fancy to the boy, who amused himself with her by
the hour at a time. Up-stairs, in the warm August morning, the two sisters
were lounging and talking "like magpies," as Jack had said when he left
them. And this is what Rob wrote:—

Dear Mother,—We got here all right. We came in a sleeping-car to

Clayton, and there we took a boat and came here. On the way we had a
good time, only a woman was mean to Fred. I paid her up, though. I will
tell you about it some day. I liked the porter on our car. I think I'd like to be
one. All you have to do is to make beds and bring drinks to people and get
them tables and black their boots, and most everybody gives you a dollar.
We had ours, supper, I mean, on a table, and it was lots of fun. Have the rats
eaten any more chickens? Island Den is a lovely house, very large, and it is
right by the water. There isn't any other house on the island, but on the next
there is a great big hotel. There are lots of islands. To-morrow cousin Alice
says I may go fishing at the end of the island. She isn't as nice as cousin
Bess, but she is pretty good. I don't think Fred likes her much. They have a
tennis court here and a boat. Has Phil come home? Puck liked the book you
sent her. She has written to tell you so. I think it is a good letter for a little
girl only five years old. Fred is in the hammock with Alice. She says, Don't
you fink boys is naughty? I hope you don't forget the worms for my turtle.
He wants five a day every day. I think this is all I can think of now. Fred
sends love, so no more now.

Your affectionate son,

ROBERT MACMILLAN ATKINSON.

P.S. I forgot to tell you that the box under my table has a worm in it that
I want to have spin himself up, so don't move it. R.M.A.

P.S. Number 2. Tell Ted I forgot to give him back his bat. It is in the
corner of the closet in my room. ROB.

P.S. 3. The best worms are in the bed where the verbenas are. R.

Folded inside this letter was another, written in large letters on a grimy
sheet of paper.[1]

Marian C. Rogers.
New York City.
Dear Aunty
Bess I want
to thank you,
for those nice
pctires you sent me.
 In the cot oer the hill,
Lives little Jennie Gill.
She is but a tot,
As big as a dot,
How do you do?
I hope that yur doll is well.
And that your dog tray is well.

[1] A genuine letter, written by a child of five.

CHAPTER XIII.

THE BOYS MEET AN OLD FRIEND.

"Help! Help! He-e-elp!"

It was a boy's voice that rang out across the waters of the Saint
Lawrence, from a dainty little rowboat that was lazily drifting down the
river. The boy was Rob. He stood up in the bow of the boat, looking to the
right and left for help; while Fred had dropped to the seat in the stern,
where he sat, white and still, waiting and listening.

"Nobody yet," said Rob, trying to speak bravely, although his tone was
far from cheerful. "We shall run across somebody soon."

"Aren't there some rapids down below here?" asked Fred anxiously.

"Ye-es," admitted Rob. "But I don't know just where they are. They're
the salt—something or other. I've heard cousin Alice tell about going
through them in a steamer. I wish I'd studied my geography a little more,
and then I'd have known how far down they are."
 This was the outcome of Rob's fishing expedition. Early that August
afternoon, he and Fred had gone down to the lower end of the island, at
some distance from the house. After Rob had fished for a half-hour, with
but poor success, he proposed to Fred that they should sit in the little green
and white boat that was drawn up on shore, and he would fish from there.
Fred fell in with the idea, and the next minute the boys were luxuriously
lounging in the stern, quite unconscious of the fact that their motions had
rocked the boat until it had left the bank and was quietly drifting off down
towards the Atlantic, with never an oar on board.

If the boys had but known it, their situation was far from alarming. It
was still quite early, so there were yet several hours of daylight before them,
and they would soon be seen and rescued. Still, it was not exactly pleasant
to be slowly moving away from home, with a very uncertain prospect of
returning in time for dinner. And added to Rob's alarm for himself was the
uncomfortable feeling that he had been the means of getting Fred into a
scrape, and that cousin Alice would wish she had not invited him to her
house.

"Boat ahoy!" called a clear voice across the water.

Rob looked around and saw a little boat with one occupant suddenly
turn from the shore, where it was creeping along in the shade, and come
darting towards them, with a long, steady sweep of the oars that told of an
experienced rower. He answered the call, and then turned to communicate
the good news to Fred, as the other boat came quickly alongside.

"Throw me your painter," said the young man who was in the boat; "I'll
take you in tow. But how did you two youngsters ever happen to get in such
a plight?"

Rob briefly explained their situation, honorably taking all the blame for
the carelessness.

"Well, never mind. You'd better come into this boat," said their rescuer.
"I can row you better that way."
 Rob carefully helped Fred to step from one boat to the other, with the
assistance of the young man, who at once noticed Fred's infirmity, and,
taking his hand, guided him to his seat in the stern, where he gazed at him
attentively, almost curiously, while Rob was seating himself by his side.

"Now," went on the stranger, when they were settled, and the other boat
made fast, "where are you boys trying to go? And where did you come
from?"

"Island Den," answered Rob. "Perhaps you don't know where that is, but
it's up by the hotel. We'll be ever so much obliged if you will take us back."

"I can do it as well as not," said their new friend. "I am on my way to
the hotel now. And I do know Island Den, for I was going to call there to-
morrow."

"Why, do you know cousin Jack?" asked Rob in astonishment.

"If cousin Jack is Mr. Rogers," said the stranger, laughing at Rob's
surprise, "I know him quite well. But how does it happen that I have never
heard of this small cousin?"

"Oh, he's no real cousin. Cousin Alice, Mrs. Rogers, is my cousin, and
I've never been here before. I'm Rob Atkinson, and I came here with cousin
Bess and Fred, this fellow, three days ago."

At the mention of these three names, a sudden idea seemed to cross the
young man's mind, and, looking closely at Fred again, he said,—

"I thought I had seen Fred before, and now I know I have."

"Yes," assented Fred quietly. "I knew your voice as soon as you and Bob
began talking. Aren't you Mr. Muir?"

"I certainly am," he answered, "and very glad to see you again. I was
sure I knew your face as soon as I saw you. And this is the Rob who tied up
the cat's feet in papers, is it?"

"Oh, Mr. Muir," began Rob, blushing at the recollection, "I didn't"—
 "Never mind that," said Mr. Muir; "but how odd that Miss Carter should
be related to Mrs. Rogers, and that I should meet her up here!"

"They're sisters," said Rob, "but cousin Alice is lots older. She's real
nice, but she isn't like cousin Bess one bit, and I don't think I like her as
well."

Fred looked horrified at Rob's alarming frankness, but Mr. Muir only
laughed, as he said,—

"I think perhaps I agree with you, Rob."

As the boat drew near the landing, no one was in sight about the piazza
or lawn of Island Den. Frank Muir pulled out his watch.

"Only half-past three now," he said, as if to himself; "still, I think I shall

risk a call, even if it is rather early, and I am not in full dress. Rob, do you
think your cousins would see me now? As long as I am all here, I think I'll
not go away without seeing them."

"Oh, I'm sure they will," said Rob confidently, as he offered his arm to
Fred, and they turned towards the house. As they came under the windows,
he called out loudly,—

"Cousin Bess, come on down here! Fred and I were carried off down the
river, and I want to tell you how we got home again."

"In just a minute, Rob," answered Bessie's voice from above.

Rob turned to his new friend with a smile of pleased anticipation.

"I thought I'd give her a surprise party," he explained, "and not tell her
you were here."

Now it happened that the day was so warm that the sisters, feeling safe
from all interruption, were lounging in Alice's room, having a long
afternoon rest before dressing for dinner. At Rob's summons, Bess hastily
twisted up her hair, put on a long wrapper of some creamy, clinging wool,
and thrust her feet into an ancient pair of slippers, whose soles and uppers
were rapidly parting company. Thus attired, she ran lightly down the stairs,
and out on the piazza, exclaiming,—

"What have you boys been"—

And then stopped aghast, as she caught sight of Mr. Muir, who rose to
meet her.

"There! I told you she'd be astonished," commented Rob triumphantly.

"Only think, cousin Bess, he found us floating off down the river, and he
knows cousin Alice and all."

A week later, Rob was waked early one morning by a sound of

splashing water. For a moment he lay in that pleasant interval between
sleeping and waking, dreamily listening to the morning twittering of the
birds, and feeling vaguely that something very pleasant was in prospect. But
an inquisitive sunbeam would shine directly into his eyes, and, as he rolled
over, he opened them to find that Fred was not in bed.

"Why, Fred, where are you?" he exclaimed.

"Here," responded a voice from the other side of the room. "I haven't
been asleep for ever so long, and my face felt so funny and hot I got up to
put some cold water on it. I don't know what's the matter, but it feels so
queer."

Rob raised his head from the pillow, and eyed his friend curiously for a
moment.

"Queer!" he said then, "I should think it might! You just ought to see
yourself, Fred Allen. It's all red and speckled—I'll tell you, you must have
hit some poison yesterday morning when we were out in the woods."

"I wonder if that is it," said Fred rather disconsolately. "My head aches
enough to have it almost anything. How long does it last, Bob?"

"Oh, two or three weeks," answered Rob encouragingly. "I've been

poisoned lots of times, and it's horrid. Pretty soon you'll begin to itch, and
then you mustn't scratch it, or it will be worse. Want me to call cousin
Bess?"

"Not now," said Fred, as he struck the repeater that his father had
bought for him soon after his return from Boston. "Only five o'clock, three
hours to breakfast time. It would be too bad to disturb her."

Rob subsided into drowsiness for a few moments, but his conscience
would not let him sleep, when he knew Fred was so uncomfortable.

"I'll tell you, Fred," he said suddenly, "they told me once, just as I was
getting over it, that plantain leaves are good for poison. You just keep quiet,
and I'll go look for some."

And he sprang out of bed and hastily pulled on his clothes, without
stopping for shoes and stockings. Out he ran, barefooted, over the dewy
lawn, looking here and there for the coveted plant. But it was not in vain
that Jack Rogers had a fine gardener for his summer home, and to the
water's edge the smooth, even turf was broken by no weed. At last, out by
the back door, Rob discovered two of the green leaves, and, seizing them in
triumph, he tiptoed up the stairs, past Bessie's door, to his own room.

"I've found two, Fred," he announced. "I've forgotten just how they said
use them, but I think it was just to put them on outside. You'd better put one
on each cheek, for they are the worst."

"How shall I make them stay?" asked Fred, after trying to balance the
smooth, slippery things on his face.

Rob pondered a moment.

"Wet them," he suggested. "That ought to make them stick."

And he crept into bed again, clothes and all, and quite regardless of the
mingled dew and dust on his small bare feet.

"I don't see why I had to go and get poisoned," said Fred, as he
thoughtfully rubbed his puffy countenance. "Just the last of the time we're
to be here, too."

"Say, Fred," asked Rob suddenly, "don't you wish we hadn't found Mr.
Muir that day?"

"I should say he found us," said Fred. "But I like him ever so much;
don't you?"

"Not very well. He's nice enough, but he's been round all the time. He
has been here every single day, and cousin Bess is always playing tennis or
going rowing with him, when I want her to do something, and— Hullo!
there goes one of your leaves." And Rob carefully replaced it on the reddest
part of Fred's face.

"Well," said Fred, "she's always ready to do things for me. Mr. Muir is
here ever so much, I know, and somebody has to entertain him; but Mrs.
Rogers is generally busy, so I suppose Miss Bess has to do it."

"I don't think she minds much," replied Rob grimly. "And last night, you
know, I told you it was bright moonlight, and they were out on the piazza
ever so long. After you went to sleep, I heard them. I don't want him round
in the way, and I am glad we are going home next week. And, you know,
Fred, she always dresses up when he comes."

"I don't see what that's for," answered Fred loyally. "She's always pretty
enough."

"Yes, I know," said Rob loftily, from the height of his thirteen years'
experience of life and its problems; "but women do that kind of thing, when
they like anybody. Say, how do you feel, Fred?"

"Horrid!" said Fred tersely.

"Didn't those leaves do any good?" inquired Rob, as he sat up in bed.

"Not yet, Bob. But I wish Miss Bess needn't know, for to-day they're all
going on that picnic up the river, and I'm afraid she won't go."
 "Can't you?" asked Rob anxiously, for as this was to be the crowning
festivity of their visit, his heart had been set on it, and ever since he had
discovered Fred's poison, he had been longing, yet fearing, to start the
subject.

"I don't feel much like it," said Fred. "I don't care at all, for picnics
aren't as much fun for me as they used to be." Here Rob gave his friend's
hand a consoling squeeze. "But you can all go and leave me, Bob. I shall be
all right, and I want you to go just the same."

When Rob entered the breakfast-room, two hours later, he said to his
cousin,—

"I wish you'd go up to Fred, a minute."

"Is anything the matter?" asked Bess, who was always anxious about
her charge.

"No, only he doesn't feel very well," answered Rob, as he followed her
out of the room. When they were alone in the hall, he went on hurriedly,
"He's poisoned a little, I think, but he doesn't feel like going to-day, and he
wants us all to go and leave him. You make him think we will, and I'll start
with you, and then, after you are gone, I'll come back to the house again. I
truly don't care about it."

Bess read her little cousin's generous motive, and as they went up the
stairs, she insisted that he should join the frolic, and let her stay; but Rob
held firm, and she had to yield, much against her will, for she knew how the
boy had anticipated the day's fun.

A striking picture met Bessie's gaze, as she went into the boys' room.
Fred had attempted to get up, as usual, but after dressing, he felt so ill and
miserable, that he had thrown himself down again. His face had swollen
until his eyes were half closed, and its ruddy hue was heightened by its
contrast with his white flannel blouse and the two bright green leaves that
Rob had again plastered on his face, just before he went down-stairs. The
remedy, applied in that way, was so original that Rob was at once dubbed
"the doctor," a name that clung to him, to his disgust, till the end of the
visit.

It was hard to see the gay party starting off in their three boats; Mr. Muir
rowing Bess in the first, Jack, Alice, and the children in a second, and the
third in charge of a servant, with a tent and the lunch. Several friends from
the hotel were to meet them, and among them was one little girl, with
whom Rob had established quite a friendship. Yes, it would be great fun,
but there was Fred, blind, ill, and alone, and the thought of his friend helped
him to smile bravely and answer decidedly all their entreaties to go.

"I think Fred doesn't need you," Bess had said. "I am glad to have you
willing to stay, Robin, but I am sure he really won't mind being alone."

"I'd rather stay," said Rob, and nothing could change his purpose.

But as the boats vanished around a point of land, Rob's resolution failed,
and for a moment his face twitched. Then he started off, and tramped twice
around the shore of the little island, as if running a race with himself. That
done, he went into Bessie's room, took a book that she was reading aloud to
Fred, and presented himself before the boy, who, now stripped of his
foliage, had settled himself for a long, dull day.

"Got left," he said briefly, as he seated himself.

And Fred understood the sacrifice.

CHAPTER XIV.

PHIL'S FIGHT.

The first of September found the boys all at home again, after their
summer fun and wanderings. Phil had been visiting his grandmother in
Vermont; Sam had gone with his family to Newport, where his boyish soul
was greatly tried by their attempts to live in a truly fashionable manner;
Bert had been in Western New York, visiting some farmer friends, who
feasted him on milk and honey, and let him go fishing and ride the horses
bareback, to his heart's content; while poor Ted was left to pine at home.
But every joy has its accompanying sorrow, and glad as they were to be
together once more, the immediate prospect of school was a cause for
mourning. To Fred, it seemed strange to hear the other five boys bemoaning
their fate, when he so wished he could go back into school again, and he
could scarcely realize that only lately he had shared their feelings. He
needed no urging to return to his pleasant lessons with Bess; but the others,
who had so many more resources, were by no means reconciled, and the
first Monday in September saw them walking slowly, very slowly, towards
the schoolhouse, with their books in their hands and rage in their hearts.

All of us who have been boys know how hard it is to leave all the frolics
and idle enjoyment of the long vacation, to sit for five hours a day in a close
room, amid the buzz of voices, and, with warm, sticky hands, turn over the
leaves of the books that never before seemed half so prosy and dull—since
last September. How all the out-door sounds that come in at the open
windows, the notes of the birds, the hum of the passing voices, the distant
bark of our own Nep or Rover, even the whir of a mowing machine in the
next yard, tempt us to throw aside the lessons, and, braving the whipping
that we know must certainly follow, to run out at the door, down the stairs,
and into the clear yellow sunshine that was surely created for boys to enjoy
themselves in! And how all the memories of the summer fun will come into
our minds, replacing the War of 1812 with a boat-race, and making the
puzzling mysteries of the binomial theorem give place to an imaginary
brook and a fish-line! Well, well! It is only what comes to us grown-up
children, when we have taken a day, or a week, or a month from our
business, and then have to settle down to work again.

One afternoon about two weeks after the opening of school, as Bess was
coming in from some errands, she found five excited boys sitting on her
front steps, eagerly waiting to see her. As she approached, she heard Rob
saying,—
 "I didn't think Phil had so much grit. If it had been you, Bert, or Sam"—

"Well, my boys," said Bess, as she sat down in the midst of them, and
took off her hat, "what is the occasion of this call? You look as if something
were the matter."

"Matter enough!" said Sam. "That Miss Witherspoon hadn't ought to

teach school anyway!" And he scowled darkly on the unconscious Fred,
who chanced to be in range of his glance.

"Sam! Sam!" remonstrated Bess.

"It's a fact, Miss Bessie," said Bert. "She's too old and cross for
anything! Just think, she's going to keep Phil after school and whip him!"

"Yes," put in Ted, "and it isn't fair."

"Phil!" said Bess incredulously. "You don't mean that Phil Cameron has
to be whipped in school! What has he done?"

"He hasn't," said Rob. "I don't think he did it at all, only she doesn't
know who did, and so she is going to whip Phil."

"Jiminy!" said Ted, rolling off the steps to the ground, in his excitement.
"I'd like to go for her! It's a burning shame to whip Phil. There isn't a better
lad in all the school, and she likes him herself, when she isn't mad."

From these remarks, however emphatic and lucid they might seem to
the boys who were in the secret, Bess had gathered but the one fact, that
Phil was in disgrace at school and was to be whipped. To her mind, corporal
punishment in schools was degrading and brutalizing, and the idea of its
being employed on a refined, gentle boy, like Phil, shocked her and roused
her indignation, for she knew the lad well enough to be sure that he had
done nothing to justify such extreme measures.

"I'll tell you about it, Miss Bess," said Bert. "You see, Phil has been
feeling funny all day, and when we marched round to get the dumb-bells, he
just turned his toes square in, and waddled along, so," and Bert illustrated
the proceeding for Bessie's benefit. "We fellows all laughed, and that rattled
Miss Witherspoon awfully, and started her down on him. I guess she didn't
feel just right to-day, perhaps. Well, by and by, when we were studying, all
of a sudden somebody snapped a great agate up the aisle, right bang against
Miss Witherspoon's desk. It astonished her and made her jump, but she
picked it up and only said, 'If this happens again, I shall whip the boy that
does it,' and then went on with her class. Pretty soon another one went
rolling along, but she wasn't quick enough to catch the boy, so she began
asking us all if we knew who did it. We were all the other side of the room
but Phil, and he was the only one in the room that said he did know. Miss
Witherspoon asked him who it was, but he just shut his mouth. Then she
asked if he did it, and he just said 'No.' And then she told him she'd whip
him unless he told, but he just wouldn't, and I say, Good for him!"

"Hurrah for Phil!" said Ted, turning a somersault on the turf.

Bess looked perplexed. She knew Miss Witherspoon too well, a veteran
teacher who had grown hard in the service, a nervous old maid who ruled
her children with an iron rod, and then went home and wept bitter tears
because they did not love her, conscientious to a fault, and at heart anxious
for the good of her pupils, although no consideration would make her take
back a hasty word, or lighten a punishment ordered in a moment of anger.
This was the first time that one of the I.I.'s had been publicly punished in
this way, and each one of them felt the disgrace as keenly as if it had been
his own, while with one consent they had come to Bess for advice and
consolation.

"There comes Phil, now!" exclaimed Rob.

Bess gave one look at the small figure coming along the street, with his
hat pulled down over his face, and his hands plunged deep into his pockets.

"I don't believe he will feel like seeing you boys now," said she. "I want
to have a little talk with him, and you had better keep away."

The boys obediently retired through the back gate before Phil had a
chance to see them. He was going directly past the house, when Bess called
him,—
 "Come in a minute, Phil."

The boy stopped doubtfully for a moment. Then he turned and came up
to where she stood waiting. Taking his hand, all red and puffed up with the
blows, she led him into the house.

"Now, Phil, my boy," she said gently, "tell me all about it."

Phil's face grew red, and his lips twitched. Then he answered abruptly,
—

"There's not much to tell, only Miss Witherspoon whipped me because I

wouldn't tell on one of the boys, and she isn't going to let me go back to
school until I tell who did it. She'll just have to wait, then, that's all."

Bess looked anxious. This was worse than she expected.

"But, Phil," she said, "isn't the boy manly enough to confess, rather than
see you suffer for him?"

Phil shook his head.

"No, he'll never tell."

"And you really had nothing to do with it?"

The boy had been sitting with his elbows on his knees and his chin in
his hands, gazing at the floor; but at this question he threw up his head
proudly, and looked straight into Bessie's eyes.

"Miss Bess," he said simply, "I told Miss Witherspoon I didn't, upon my
honor, and did you ever know me to lie?"

"No, Phil, I never did."

"I think she might believe me, too, then," muttered Phil, as he settled
back after his momentary flash. "She thinks I did it, and won't believe me
when I say I didn't. Oh, how I hate to tell my father!" And he started up to
go.
 "Will you tell me, Phil, who it was?" asked Bess, as she followed him to
the door.

Phil shook his head again.

"But I might be able to straighten the matter out. You mustn't lose your
school."

"I'll lose it always, rather than be a tell-tale."

The boys were loud in their exclamations when they heard, the next
morning, that Phil was suspended from school. One after another, they
coaxed, wheedled, begged, and stormed by turns, but Phil could not be
induced to tell them his secret, although one word would have put him back
in his classes again. At Bessie's suggestion, Fred urged Phil to tell him, as
long as he was outside the school set, but it did no more good than Bessie's
call did on Miss Witherspoon.

"Yes, I am sorry," that worthy woman confessed; "I was tired that day,
and I think I was hasty, for I don't think Philip is a bad boy at heart. It was a
little thing to punish so severely, but, if I give in now, I shall lose all my
control for the future. Let the boys once feel that they can make me yield,
and I might as well give up teaching."

Poor Miss Witherspoon! After all her years of teaching, she had yet to
learn how quickly all pupils respect a teacher who can make herself as a
little child in acknowledging a mistake, and making what reparation for it
she can.

But a week had passed, and Phil was as obstinate on one side as his
teacher was determined on the other. In vain his father and mother urged
and commanded. Angry and smarting from the injustice done him, this
seemed a different Phil from the pleasant, happy-go-lucky lad they used to
know. At length, Mr. and Mrs. Cameron, at their wits' end, begged Bessie to
take Phil in hand.

"Oh, dear!" Bess said to her mother, on the evening after this
remarkable request. "I do wish people would discipline their own children.
The idea of expecting me to succeed where they fail! It is too absurd."

However, Phil was invited to dine at the Carters', whither he went

somewhat suspiciously, for he regarded this as only a new plot to entrap
him into telling what he had made up his mind to keep to himself. But Bess
was wily. Dinner-time came and went, and no word of the dreaded subject,
until Phil began to think that his had been a false alarm. But by and by Mrs.
Carter had gone out of the room, and Fred went away in search of Fuzz.
Then Bess moved a chair up before the open fire, and pulled a low stool to
its side.

"Come, Phil, I want to talk."

Phil obediently settled himself at Bessie's feet, and prepared for the
worst; but Bess only began to talk about the boys and the club. The child
was just congratulating himself on his continued escape, when she suddenly
asked,—

"What do you think I have started the club for?"

"I don't know. Fun, I suppose."

"Partly for that, but, still more, to improve us in all sorts of ways. And
yet I find I have failed to teach you the very first lesson of all."

"What's that?" asked Phil curiously.

"Obedience, Phil. Your father and mother wish you to tell Miss
Witherspoon who threw that marble, and you refuse to obey them."

"I'm not going to tell tales," said Phil sullenly.

Bess rested her hand lightly on the smooth brown head.

"Phil, the first duty you have now is to be guided by your father and
mother. They know so much better than you what is right for you. I can see
how hard it is for you to give in, in this case. But while a sneak and a tell-
tale is the meanest of boys, you would not be either, under these
circumstances."
 "Yes, I should," answered Phil. "It's a mean thing to do, and the fellows
would all be down on me."

"Suppose they were?" replied Bess. "Is it your parents or 'the fellows'
that you want to please? I will tell you what one trouble is, Phil; you have
read too many stories where the hero nobly bears the punishment for
another boy, and is only cleared on the last half-page. Isn't it true?"

Phil laughed, in spite of himself.

"That would be all very well if you had no duty to any one but yourself;
but, back of that, you owe obedience to your father and mother, and if they
think that you ought to go back into school, that is what you should do. You
are too young, my boy, to decide these things for yourself. And it is because
we have so many hopes and plans for your future that we want you to do
right now, every day. It will be hard for you to go back, but, even if it is, we
all want you to go. Will you promise?"

Phil's face had softened at her last words.

"I won't promise, Miss Bess, for then I should have to, anyway, and I'm
not sure yet, till I think it all over. I'll tell you to-morrow."

Bess patted his shoulder approvingly, for this was a concession at least.
Then she went on, after a little pause,—

"Phil, dear, ever so long ago, Fred and I took for our motto a verse from
your All Saints' Hymn,—'Oh, may thy soldiers,' and we are trying to win
our 'victor's crown.' Why not take it for your motto, too? You boys all have
a good deal of the stuff that makes heroes and fighters. Just now you are
forgetting that a soldier's first duty is to obey his superior officer, and that
any disobedience, even a slight one, may ruin the whole campaign. Will this
small soldier join our company, and fight with us, 'faithful, true, and bold'?"

"Ye-es, I s'pose so."

"Even when you remember that your first step must be to yield your
idea of right to your father's?"
 "Ye-e-es."

It was a long-drawn yes, and it told of a whole battle, and a victory. As

Bessie bent over the boy for a moment, she saw that the lashes over the
gray eyes were a little damp, and the lips were quivering. But there was no
time for Phil to have so much as a tear, for just then the door opened and
Ted rushed in, capering like a mad creature, while Fred stood beaming in
the doorway.

"Why, Ted, what is the matter?" exclaimed Bess in wonder, as Ted

rushed up to Phil, shook both hands furiously, and then backed out into the
middle of the room, where he executed a sort of clog-dance, to the rage of
Fuzz, who barked himself hoarse, from the shelter of his basket, whither he
had retired for safety.

"Jack Bradley fired that marble!" said Ted, interrupting his antics for a
moment, and then resuming them again more vehemently than ever, while
Fuzz leaped from his basket and rushed distractedly this way and that,
adding his voice to the general confusion.

"How do you know?" asked Bess, although a glance at Phil's face was
enough to assure her that Ted's statement was true.

"I'll tell you," said Ted, composing himself as well as he could on such
short notice, while Fred deliberately seated himself in the place lately
vacated by Phil.

"You see," he began, "we boys have all been mad about Phil's scrape,
and we have just formed a regular league of detectives. This is the way I
went to work. That marble came out of Phil's aisle. Well, it came up out of it
sort of cornerwise, and bounced off the other way. That showed the
direction, so I was pretty sure which side of the aisle it started from. Then,
half-way down the aisle is where that little milksop of a Jimmy Harris sits.
He never could tell a lie, just like Washington—don't believe he knows
enough! But he's always looking round, and would have seen who fired it, if
it had been anybody in front of him, so I made up my mind 'twasn't. Then I
knew it must have been one of three boys, so I went to work. I kind of
suspected 'twas Jack; he's a mean lad, anyhow. So yesterday I began to talk
about Phil to him, and he was very talky, said 'twas a mean shame and all
that, but he never once looked me in the eye. Thinks I, 'I don't believe you.'
Then I asked Miss Witherspoon to let me see the agate. It was a queer one,
and after school I went the rounds of the stores, looking for some like it. I
found a whole lot at Smith's, and they told me they had just come in new
last week. I said I thought I would take one or two, and get the start of all
the boys; but the clerk said I was too late, for Jack had bought some the
other day. That clinched the matter, for they were different from any I ever
saw. I don't believe Jack knew he had that one in his hand, or he wouldn't
have fired it. He's too stingy. Well, to-night after school, I asked him if he
wanted to swap marbles. He looked rather uncomfortable, and said he
hadn't had any since last spring. I asked him how about the ones he had just
bought of Smith. He just turned all colors, and begged me not to tell, for
he'd get a whipping, and another at home. Great baby! But I didn't tell. I just
gripped my arms round him, and hauled him up to Miss Witherspoon, and
told her to ask him about Phil and the marbles; that's all. I had to carry the
milk, so I couldn't go to Phil's till just now, and, when I found he was here, I
came right after him. And he can go into school in the morning and— Oh,
jiminy—scratch!"

There was a crash. Ted, always in perpetual motion, in his present

excitement had seated himself sideways in a low rocking-chair, and with
one hand on the back, the other clutching the edge of the seat, he had been
rocking furiously to and fro, till at this point he went a little too far, and,
losing his balance, he landed in an ignominious pile on the floor, amid the
shouts of the other two boys.

CHAPTER XV.

"GREATER LOVE HATH NO MAN."

 It was one of Fred's blue days; for, though they came more rarely, there
were often times when his trouble seemed more than he could endure, and
he was either irritable and moody, or so sad and despondent that even Bess
was in despair over him. For a long time he had been brave and bright, but
now the reaction seemed to have set in, and on this particular day he was
harder to manage than usual. The other boys had all gone away to a ball
game, to which they had tried in vain to induce Fred to accompany them.
Of late he had gone about with them to many of their frolics, but to-day he
had refused to join them. He was lying in a hammock out in the warm
midday sun of late September, and feeling at war with all the world but
Fuzz, who lay curled up across his breast with his head laid on the boy's
shoulder, occasionally nestling about a little, or giving a lazy growl
whenever Fred ventured to move.

Out on the lawn, Bess and Mr. Muir were playing tennis,—for it was
strange how often the young man had occasion to spend two or three days
with Mr. Washburn. Fred could hear the thud of their balls against the
rackets, and listened idly to their voices; but although his admiration for Mr.
Muir amounted to a sort of hero-worship, he was too cross and dismal to-
day to follow him about, as he usually did, or to respond to his pleasant,
merry greeting. Everybody was having a good time but just himself, and he
couldn't do anything at all. Everything was going wrong to-day. Miss Bess
was too busy to read to him, just because that bothering old Mr. Muir was
always round,—and, for a moment, Fred almost hated his idol. If he had
only known that he was going to be here, he would have gone with the
boys. He wished he had.

Fred's meditations had just reached this point, when he heard Rob's
voice calling from the street,—

"Cousin Bess, where's Fred?"

"In the hammock, Rob. What sends you home so early?"

"Early!" thought Rob mischievously, "I've been gone nearly three hours
and a half. Mr. Muir must be exciting, if time goes so fast with him round."
But all he said was,—