Download the full version of the textbook now at textbookfull.

com

Intelligent Systems and Applications:

Proceedings of the 2020 Intelligent Systems
Conference (IntelliSys) Volume 2 Kohei Arai

https://textbookfull.com/product/intelligent-
systems-and-applications-proceedings-of-
the-2020-intelligent-systems-conference-
intellisys-volume-2-kohei-arai/

Explore and download more textbook at https://textbookfull.com

 Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.

Intelligent Systems and Applications: Proceedings of the

2020 Intelligent Systems Conference (IntelliSys) Volume 3
Kohei Arai
https://textbookfull.com/product/intelligent-systems-and-applications-
proceedings-of-the-2020-intelligent-systems-conference-intellisys-
volume-3-kohei-arai/
textbookfull.com

Intelligent Computing Proceedings of the 2020 Computing

Conference Volume 2 Kohei Arai

https://textbookfull.com/product/intelligent-computing-proceedings-of-
the-2020-computing-conference-volume-2-kohei-arai/

textbookfull.com

Intelligent Computing Proceedings of the 2020 Computing

Conference Volume 3 Kohei Arai

https://textbookfull.com/product/intelligent-computing-proceedings-of-
the-2020-computing-conference-volume-3-kohei-arai/

textbookfull.com

Global Sourcing of Digital Services Micro and Macro

Perspectives 11th Global Sourcing Workshop 2017 La Thuile
Italy February 22 25 2017 Revised Selected Papers 1st
Edition Ilan Oshri
https://textbookfull.com/product/global-sourcing-of-digital-services-
micro-and-macro-perspectives-11th-global-sourcing-workshop-2017-la-
thuile-italy-february-22-25-2017-revised-selected-papers-1st-edition-
ilan-oshri/
textbookfull.com
Spirituality in Nursing: Standing on Holy Ground 6th
Edition Mary Elizabeth O'Brien

https://textbookfull.com/product/spirituality-in-nursing-standing-on-
holy-ground-6th-edition-mary-elizabeth-obrien/

textbookfull.com

Vampire With Benefits A Mythmatched story Supernatural

Selection Book 2 E.J. Russell

https://textbookfull.com/product/vampire-with-benefits-a-mythmatched-
story-supernatural-selection-book-2-e-j-russell/

textbookfull.com

Art Alphabets Monograms and Lettering First Edition J M

Bergling

https://textbookfull.com/product/art-alphabets-monograms-and-
lettering-first-edition-j-m-bergling/

textbookfull.com

Active Geophysical Monitoring 2nd Edition Junzo Kasahara

(Editor)

https://textbookfull.com/product/active-geophysical-monitoring-2nd-
edition-junzo-kasahara-editor/

textbookfull.com

Excel Simple and Effective Strategies to Learn and Execute

Excel Programming Volume 3 3rd Edition Mr Daniel Jones

https://textbookfull.com/product/excel-simple-and-effective-
strategies-to-learn-and-execute-excel-programming-volume-3-3rd-
edition-mr-daniel-jones/
textbookfull.com
Think Trade Like a Champion The Secrets Rules Blunt Truths
of a Stock Market Wizard First Edition Mark Minervini

https://textbookfull.com/product/think-trade-like-a-champion-the-
secrets-rules-blunt-truths-of-a-stock-market-wizard-first-edition-
mark-minervini/
textbookfull.com
Advances in Intelligent Systems and Computing 1251

Kohei Arai
Supriya Kapoor
Rahul Bhatia Editors

Intelligent
Systems and
Applications
Proceedings of the 2020 Intelligent
Systems Conference (IntelliSys)
Volume 2
Advances in Intelligent Systems and Computing

Volume 1251

Series Editor
Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences,
Warsaw, Poland

Advisory Editors
Nikhil R. Pal, Indian Statistical Institute, Kolkata, India
Rafael Bello Perez, Faculty of Mathematics, Physics and Computing,
Universidad Central de Las Villas, Santa Clara, Cuba
Emilio S. Corchado, University of Salamanca, Salamanca, Spain
Hani Hagras, School of Computer Science and Electronic Engineering,
University of Essex, Colchester, UK
László T. Kóczy, Department of Automation, Széchenyi István University,
Gyor, Hungary
Vladik Kreinovich, Department of Computer Science, University of Texas
at El Paso, El Paso, TX, USA
Chin-Teng Lin, Department of Electrical Engineering, National Chiao
Tung University, Hsinchu, Taiwan
Jie Lu, Faculty of Engineering and Information Technology,
University of Technology Sydney, Sydney, NSW, Australia
Patricia Melin, Graduate Program of Computer Science, Tijuana Institute
of Technology, Tijuana, Mexico
Nadia Nedjah, Department of Electronics Engineering, University of Rio de Janeiro,
Rio de Janeiro, Brazil
Ngoc Thanh Nguyen , Faculty of Computer Science and Management,
Wrocław University of Technology, Wrocław, Poland
Jun Wang, Department of Mechanical and Automation Engineering,
The Chinese University of Hong Kong, Shatin, Hong Kong
The series “Advances in Intelligent Systems and Computing” contains publications
on theory, applications, and design methods of Intelligent Systems and Intelligent
Computing. Virtually all disciplines such as engineering, natural sciences, computer
and information science, ICT, economics, business, e-commerce, environment,
healthcare, life science are covered. The list of topics spans all the areas of modern
intelligent systems and computing such as: computational intelligence, soft comput-
ing including neural networks, fuzzy systems, evolutionary computing and the fusion
of these paradigms, social intelligence, ambient intelligence, computational neuro-
science, artificial life, virtual worlds and society, cognitive science and systems,
Perception and Vision, DNA and immune based systems, self-organizing and
adaptive systems, e-Learning and teaching, human-centered and human-centric
computing, recommender systems, intelligent control, robotics and mechatronics
including human-machine teaming, knowledge-based paradigms, learning para-
digms, machine ethics, intelligent data analysis, knowledge management, intelligent
agents, intelligent decision making and support, intelligent network security, trust
management, interactive entertainment, Web intelligence and multimedia.
The publications within “Advances in Intelligent Systems and Computing” are
primarily proceedings of important conferences, symposia and congresses. They
cover significant recent developments in the field, both of a foundational and
applicable character. An important characteristic feature of the series is the short
publication time and world-wide distribution. This permits a rapid and broad
dissemination of research results.
** Indexing: The books of this series are submitted to ISI Proceedings,
EI-Compendex, DBLP, SCOPUS, Google Scholar and Springerlink **

More information about this series at http://www.springer.com/series/11156

Kohei Arai Supriya Kapoor
• •

Rahul Bhatia
Editors

Intelligent Systems
and Applications
Proceedings of the 2020 Intelligent Systems
Conference (IntelliSys) Volume 2

123
Editors
Kohei Arai Supriya Kapoor
Saga University The Science and Information
Saga, Japan (SAI) Organization
Bradford, West Yorkshire, UK
Rahul Bhatia
The Science and Information
(SAI) Organization
Bradford, West Yorkshire, UK

ISSN 2194-5357 ISSN 2194-5365 (electronic)

Advances in Intelligent Systems and Computing
ISBN 978-3-030-55186-5 ISBN 978-3-030-55187-2 (eBook)
https://doi.org/10.1007/978-3-030-55187-2
© Springer Nature Switzerland AG 2021
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Editor’s Preface

This book contains the scientific contributions included in the program of the
Intelligent Systems Conference (IntelliSys) 2020, which was held during September
3–4, 2020, as a virtual conference. The Intelligent Systems Conference is a pres-
tigious annual conference on areas of intelligent systems and artificial intelligence
and their applications to the real world.
This conference not only presented state-of-the-art methods and valuable
experience from researchers in the related research areas, but also provided the
audience with a vision of further development in the fields. We have gathered a
multi-disciplinary group of contributions from both research and practice to discuss
the ways how intelligent systems are today architectured, modeled, constructed,
tested and applied in various domains. The aim was to further increase the body of
knowledge in this specific area by providing a forum to exchange ideas and discuss
results.
The program committee of IntelliSys 2020 represented 25 countries, and authors
submitted 545 papers from 50+ countries. This certainly attests to the widespread,
international importance of the theme of the conference. Each paper was reviewed
on the basis of originality, novelty and rigorousness. After the reviews, 214 were
accepted for presentation, out of which 177 papers are finally being published in the
proceedings.
The conference would truly not function without the contributions and support
received from authors, participants, keynote speakers, program committee mem-
bers, session chairs, organizing committee members, steering committee members
and others in their various roles. Their valuable support, suggestions, dedicated
commitment and hard work have made the IntelliSys 2020 successful. We warmly
thank and greatly appreciate the contributions, and we kindly invite all to continue
to contribute to future IntelliSys conferences.

v
vi Editor’s Preface

It has been a great honor to serve as the General Chair for the IntelliSys 2020 and
to work with the conference team. We believe this event will certainly help further
disseminate new ideas and inspire more international collaborations.
Kind Regards,

Kohei Arai
Conference Chair
Contents

CapsNet vs CNN: Analysis of the Effects of Varying Feature

Spatial Arrangement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Ugenteraan Manogaran, Ya Ping Wong, and Boon Yian Ng
Improved 2D Human Pose Tracking Using Optical
Flow Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Aleksander Khelvas, Alexander Gilya-Zetinov, Egor Konyagin,
Darya Demyanova, Pavel Sorokin, and Roman Khafizov
Transferability of Fast Gradient Sign Method . . . . . . . . . . . . . . . . . . . . 23
Tamás Muncsan and Attila Kiss
Design of an Automatic System to Determine the Degree
of Progression of Diabetic Retinopathy . . . . . . . . . . . . . . . . . . . . . . . . . 35
Hernando González, Carlos Arizmendi, and Jessica Aza
Adaptive Attention Mechanism Based Semantic Compositional
Network for Video Captioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Zhaoyu Dong, Xian Zhong, Shuqin Chen, Wenxuan Liu, Qi Cui,
and Luo Zhong
Estimated Influence of Online Management Tools on Team
Management Based on the Research with the Use of the System
of Organizational Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Olaf Flak
Liveness Detection via Facial Expressions Queue . . . . . . . . . . . . . . . . . . 73
Bat-Erdene Batsukh
Java Based Application Development for Facial Identification
Using OpenCV Library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Askar Boranbayev, Seilkhan Boranbayev, and Askar Nurbekov

vii
viii Contents

Challenges in Face Recognition Using Machine Learning

Algorithms: Case of Makeup and Occlusions . . . . . . . . . . . . . . . . . . . . . 86
Natalya Selitskaya, Stanislaw Sielicki, and Nikolaos Christou
The Effects of Social Issues and Human Factors on the Reliability
of Biometric Systems: A Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Mohammadreza Azimi and Andrzej Pacut
Towards Semantic Segmentation Using Ratio Unpooling . . . . . . . . . . . . 111
Duncan Boland and Hossein Malekmohamadi
Adaptive Retraining of Visual Recognition-Model in Human
Activity Recognition by Collaborative Humanoid Robots . . . . . . . . . . . . 124
Vineet Nagrath, Mossaab Hariz, and Mounim A. El Yacoubi
A Reasoning Based Model for Anomaly Detection
in the Smart City Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
Patrick Hammer, Tony Lofthouse, Enzo Fenoglio, Hugo Latapie,
and Pei Wang
Document Similarity from Vector Space Densities . . . . . . . . . . . . . . . . . 160
Ilia Rushkin
Food Classification for Inflammation Recognition Through
Ingredient Label Analysis: A Real NLP Case Study . . . . . . . . . . . . . . . 172
Stefano Campese and Davide Pozza
Classification Based Method for Disfluencies Detection
in Spontaneous Spoken Tunisian Dialect . . . . . . . . . . . . . . . . . . . . . . . . 182
Emna Boughariou, Younès Bahou, and Lamia Hadrich Belguith
A Comprehensive Methodology for Evaluating Conversation-Based
Interfaces to Relational Databases (C-BIRDs) . . . . . . . . . . . . . . . . . . . . 196
Majdi Owda, Amani Yousef Owda, and Fathi Gasir
Disease Normalization with Graph Embeddings . . . . . . . . . . . . . . . . . . . 209
D. Pujary, C. Thorne, and W. Aziz
Quranic Topic Modelling Using Paragraph Vectors . . . . . . . . . . . . . . . . 218
Menwa Alshammeri, Eric Atwell, and Mhd Ammar Alsalka
Language Revitalization: A Benchmark for Akan-to-English
Machine Translation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Kingsley Nketia Acheampong and Nathaniel Nii Oku Sackey
A Machine Learning Platform for NLP in Big Data . . . . . . . . . . . . . . . 245
Mauro Mazzei
Contents ix

Recent News Recommender Using Browser’s History . . . . . . . . . . . . . . 260

Samer Sawalha and Arafat Awajan
Building a Wikipedia N-GRAM Corpus . . . . . . . . . . . . . . . . . . . . . . . . . 277
Jorge Ramón Fonseca Cacho, Ben Cisneros, and Kazem Taghva
Control Interface of an Automatic Continuous Speech Recognition
System in Standard Arabic Language . . . . . . . . . . . . . . . . . . . . . . . . . . 295
Brahim Fares Zaidi, Malika Boudraa, Sid-Ahmed Selouani,
Mohammed Sidi Yakoub, and Ghania Hamdani
Emotion Detection Throughout the Speech . . . . . . . . . . . . . . . . . . . . . . 304
Manuel Rodrigues, Dalila Durães, Ricardo Santos, and Cesar Analide
Understanding Troll Writing as a Linguistic Phenomenon . . . . . . . . . . . 315
Sergei Monakhov
Spatial Sentiment and Perception Analysis of BBC News Articles
Using Twitter Posts Mining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
Farah Younas and Majdi Owda
Human-Machine Interaction for Improved Cybersecurity Named
Entity Recognition Considering Semantic Similarity . . . . . . . . . . . . . . . 347
Kazuaki Kashihara, Jana Shakarian, and Chitta Baral
Predicting University Students’ Public Transport Preferences
for Sustainability Improvement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
Ali Bakdur, Fumito Masui, and Michal Ptaszynski
Membrane Clustering Using the PostgreSQL Database
Management System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
Tamás Tarczali, Péter Lehotay-Kéry, and Attila Kiss
STAR: Spatio-Temporal Prediction of Air Quality Using
a Multimodal Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
Tien-Cuong Bui, Joonyoung Kim, Taewoo Kang, Donghyeon Lee,
Junyoung Choi, Insoon Yang, Kyomin Jung, and Sang Kyun Cha
Fair Allocation Based Soft Load Shedding . . . . . . . . . . . . . . . . . . . . . . . 407
Sarwan Ali, Haris Mansoor, Imdadullah Khan, Naveed Arshad,
Safiullah Faizullah, and Muhammad Asad Khan
VDENCLUE: An Enhanced Variant of DENCLUE Algorithm . . . . . . . 425
Mariam S. Khader and Ghazi Al-Naymat
Detailed Clustering Based on Gaussian Mixture Models . . . . . . . . . . . . 437
Nikita Andriyanov, Alexander Tashlinsky, and Vitaly Dementiev
Smartphone Applications Developed to Collect Mobility Data:
A Review and SWOT Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
Cristina Pronello and Pinky Kumawat
Visit https://textbookfull.com
now to explore a rich
collection of eBooks, textbook
and enjoy exciting offers!
x Contents

A Novel Approach for Heart Disease Prediction Using Genetic

Algorithm and Ensemble Classification . . . . . . . . . . . . . . . . . . . . . . . . . 468
Indu Yekkala and Sunanda Dixit
An Improved Algorithm for Fast K-Word Proximity Search Based
on Multi-component Key Indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
Alexander B. Veretennikov
A Feedback Integrated Web-Based Multi-Criteria Group Decision
Support Model for Contractor Selection Using Fuzzy Analytic
Hierarchy Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
Abimbola H. Afolayan, Bolanle A. Ojokoh, and Adebayo O. Adetunmbi
AIS Ship Trajectory Clustering Based
on Convolutional Auto-encoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529
Taizheng Wang, Chunyang Ye, Hui Zhou, Mingwang Ou, and Bo Cheng
An Improved Q-Learning Algorithm for Path Planning
in Maze Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 547
Shimin Gu and Guojun Mao
Automatic Classification of Web News: A Systematic
Mapping Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 558
Mauricio Pandolfi-González, Christian Quesada-López,
Alexandra Martínez, and Marcelo Jenkins
Big Data Clustering Using MapReduce Framework: A Review . . . . . . . 575
Mariam S. Khader and Ghazi Al-Naymat
A Text Extraction-Based Smart Knowledge Graph Composition
for Integrating Lessons Learned During the Microchip Design . . . . . . . 594
Hasan Abu Rasheed, Christian Weber, Johannes Zenkert, Peter Czerner,
Roland Krumm, and Madjid Fathi
Clustering Approach to Topic Modeling in Users Dialogue . . . . . . . . . . 611
E. Feldina and O. Makhnytkina
Knowledge-Based Model for Formal Representation of Complex
System Visual Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 618
Andrey I. Vlasov, Ludmila V. Juravleva, and Vadim A. Shakhnov
Data Mining Solutions for Direct Marketing Campaign . . . . . . . . . . . . . 633
Torubein Fawei and Duke T. J. Ludera
A Review of Phishing URL Detection Using Machine
Learning Classifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 646
Sajjad Jalil and Muhammad Usman
Contents xi

Data Mining and Machine Learning Techniques for Bank

Customers Segmentation: A Systematic Mapping Study . . . . . . . . . . . . 666
Maricel Monge, Christian Quesada-López, Alexandra Martínez,
and Marcelo Jenkins
Back to the Past to Charter the Vinyl Electronic Market:
A Data Mining Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 685
Sara Lousão, Pedro Ramos, and Sérgio Moro
Learning a Generalized Matrix from Multi-graphs Topologies
Towards Microservices Recommendations . . . . . . . . . . . . . . . . . . . . . . . 693
Ilias Tsoumas, Chrysostomos Symvoulidis, and Dimosthenis Kyriazis
Big Data in Smart Infrastructure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 703
Will Serrano
Academic Articles Recommendation Using
Concept-Based Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 733
Dina Mohamed, Ayman El-Kilany, and Hoda M. O. Mokhtar
Self-organising Urban Traffic Control on Micro-level Using
Reinforcement Learning and Agent-Based Modelling . . . . . . . . . . . . . . 745
Stefan Bosse
The Adoption of Electronic Administration by Citizens: Case
of Morocco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 765
Fadwa Satry and Ez-zohra Belkadi

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 781

 CapsNet vs CNN: Analysis of the Effects
of Varying Feature Spatial Arrangement

Ugenteraan Manogaran(B) , Ya Ping Wong, and Boon Yian Ng

Faculty of Computing and Informatics, Multimedia University,

Persiaran Multimedia, 63100 Cyberjaya, Selangor, Malaysia
m.ugenteraan15@gmail.com, ypwong@mmu.edu.my

Abstract. Despite the success over the recent years, convolutional neu-
ral network (CNN) has a major limitation of the inability to retain spa-
tial relationship between learned features in deeper layers. Capsule net-
work with dynamic routing (CapsNet) was introduced in 2017 with a
speculation that CapsNet can overcome this limitation. In our research,
we created a suitable collection of datasets and implemented a simple
CNN model and a CapsNet model with similar complexity to test this
speculation. Experimental results show that both the implemented CNN
and CapsNet models have the ability to capture the spatial relationship
between learned features. Counterintuitively, our experiments show that
our CNN model outperforms our CapsNet model using our datasets. This
implies that the speculation does not seem to be entirely correct. This
might be due to the fact that our datasets are too simple, hence requir-
ing a simple CNN model. We further recommend future research to be
conducted using deeper models and more complex datasets to test the
speculation.

Keywords: CapsNet · Convolutional Neural Network · Spatial

relationship

1 Introduction

Ever since Krizhevsky et al. [1] demonstrated the outstanding performance of

a convolutional neural network (CNN) model on ImageNet, CNN has become
the center of attraction for computer vision researchers to solve problems such
as image segmentation, object detection, object localization, image classifica-
tion, and image retrieval. Some of the well-known CNN models are AlexNet [1],
GoogleNet [2], VGG-16 [3], YOLO [4], and RCNN [5].
A significant advantage of CNN is its ability to maintain translation invari-
ance for feature detection [6]. This means that the position of an object that
is known by CNN in the input image does not affect the performance of CNN.
CNN is able to recognize the object regardless of its position in the image. This
is achieved by the use of pooling layers in CNN which are also responsible for
reducing the size of feature maps.
c Springer Nature Switzerland AG 2021
K. Arai et al. (Eds.): IntelliSys 2020, AISC 1251, pp. 1–9, 2021.
https://doi.org/10.1007/978-3-030-55187-2_1
2 U. Manogaran et al.

However, this advantage of CNN inevitably leads to its drawbacks. Two

major drawbacks are the lack of rotational invariance [8] and the failure to retain
spatial relationships between features [7]. The failure to be invariant to rotations
would cause CNN to produce false negatives when an object that is known by
the network is rotated to a certain extent. On the other hand, the failure of
CNN to retain spatial relationships between features would cause the network
to produce false positives [9]. Even though CNNs can achieve state-of-the-art
results on many challenges despite these drawbacks, the drawbacks can become
serious concerns in applications such as in security systems.
To overcome the lack of rotational invariance, augmenting the training data
by rotations became a standard practice in the training of CNN models [10].
However, the training time would also increase tremendously. Therefore, in order
to solve the drawbacks of CNN, Sabour, S. et al. [11] proposed a novel neural
network architecture known as capsule network with dynamic routing (CapsNet).
Unlike CNN, CapsNet produces outputs in the form of vectors instead of scalars.
This allows CapsNet to retain the properties of an object such as rotation,
skewness, and thickness.
It has also been explicitly speculated in a number of research papers [9–11],
that CapsNet would be able to retain the spatial relationships between features
in contrast to CNN. In other words, features of an object in the wrong places
such as a face with eyes below the nose instead of above the nose would still be
a face to CNN. However, CapsNet is speculated to be able to identify it as a
non-face.
To the best of our knowledge, there has been no research literature that
has produced comprehensive experiments and analysis on this speculation. In
this paper we present our experiments and analysis to test both CapsNet and
CNN on this speculation. We generated our own dataset for the purpose of our
experiment and implemented a CapsNet and a CNN model. Evaluation of these
models were done on the generated dataset in such a way that the speculation
is tested.

2 Related Works

The concept of CNN was first proposed by LeCun et al. [12] in 1989. However,
due to the lack of computational power and availability of dataset, it was not
until recent years that researchers are able to develop feasible models utilizing
modern high-performance computers. One notable breakthrough was the work of
Krizhevsky et al. [1] which achieved state-of-the-art performance in the ImageNet
challenge [13] in 2012. Since then, countless researches have been conducted to
develop more advanced CNN models to be used successfully in real-world appli-
cations such as speech recognition [14], gait recognition [15], steering controls in
self-driving cars [16], human crowd detection [17], and medical image segmenta-
tion [18].
Despite successful demonstrations of CNN models, one of the pioneers, Geof-
frey Hinton argued that the current CNNs “are misguided in what they are
 CapsNet vs CNN 3

trying to achieve” [19], due to the use of pooling layers for subsampling in CNN
models. The models lose the ability to compute precise spatial relationships
between learned features in the deeper layers. When a pooling layer is used in
between convolutional layers, only the most active neuron in a local region of
a feature map would be retained while the rest of the neurons are disregarded.
Such disregard of neurons causes the loss of spatial information of the features.
Furthermore, due to the use of scalars instead of vectors, properties of fea-
tures such as orientation, thickness, and skewness are lost. Therefore, Hinton,
G. E. et al. [19] proposed to group neurons together as vectors and use them to
represent the features of an object. These vectors are called capsules.
In 2017, Hinton and his team [11] proposed an architecture called capsule
networks with dynamic routing (CapsNet) that performed better than CNN
on the MNIST dataset. It achieved state-of-the-art results in MNIST with only
0.25% of test error. The CapsNet model which achieved 99.23% on the expanded
MNIST set were able to reach an accuracy of 79% on affNIST test set while a
CNN that achieved 99.22% accuracy on the expanded MNIST test set only
achieved 66% accuracy on affNIST test set. This proves that CapsNet is more
robust to affine transformations.
We have implemented CapsNet based on the original research paper [11] and
through the reconstruction network as mentioned in the paper, it can be seen
that CapsNet preserves the properties of the features well as shown in Fig. 1.

Fig. 1. Original (Top Row) vs Reconstructed (Bottom Row) images.

Following the success of CapsNet on MNIST, studies have been conducted to

push the capability of CapsNet. LaLonde, R. et al. [20] proposed a novel archi-
tecture called SegCaps which is based on CapsNet to perform image segmen-
tation. SegCaps outperformed previous state-of-the-art networks even though
SegCaps had lower number of parameters on the LUNA16 subset of the LIDC-
IDRI database. In a different research [7], a CapsNet model outperformed a CNN
model with similar complexity in Human Action Recognition task on KTH and
UFC-sports dataset.
One of the speculated properties of a CapsNet model is its ability to retain
spatial relationships between learned feature unlike a CNN model [9–11]. In
other words, the relative positions of features are insignificant to a CNN model.
This causes a CNN model to produce false positives such as labelling an image
of a face with eyes below the nose as a face. In contrast to CNN, CapsNet is
4 U. Manogaran et al.

speculated to be able to avoid such false positives. However, to date, no studies

have been conducted to test this speculation. In this paper, we seek to test this
speculation to gain deeper insights into CapsNet.

3 Methodology
We implemented a CNN model and a CapsNet model for this study. In general, a
CNN model consists of several convolutional layers with pooling layers between
them followed by fully-connected layers as explained in Sect. 3.2. A CapsNet
model consists of several convolutional layers without any pooling layers followed
by a primary capsule layer and a digit capsule layer as explained in Sect. 3.3.
Both of the models were designed to have the same number of layers in order
for them to be comparable.
In order to test the speculation, we need to design a dataset in such a way
that there are two classes of images containing the same features but the features
from different classes have different spatial arrangements. Training our models
directly on such a dataset may not yield any insight into the models as the
models will learn to identify the shape of the entire objects successfully instead
of the distinct features as intended.
Therefore, we prepared two groups of datasets whereby the first group con-
tains images of only distinct features while the second group contains objects
formed by the composition of the features. Our models are first trained on the
dataset from Group 1. Once the training is completed, the weights of the con-
volutional layers in both models will be frozen while the weights of the rest of
the layers will be re-trained on the dataset from Group 2. This will ensure that
our models learn the distinct features first before learning to identify the objects
using the learned features. This strategy is known as transfer learning.
Below we describe in detail regarding the dataset generation, testing of con-
volutional neural network model, and testing of capsule network with dynamic
routing model. Since our datasets only consist of objects with simple features,
relatively simple models should be sufficient to achieve good accuracy on the
evaluations.

3.1 Dataset Generation

Our dataset consists of two groups. Figure 2 shows samples from Group 1, which
contains images of arrows and non-arrows. Figure 3 shows samples from Group
2, which contains images of equilateral triangles and rectangles. Each image is
of 64 × 64 pixels.
We chose to use generated images in our dataset because there is too much
ambiguity in real-life images. Furthermore, simple polygon objects were chosen as
they are well-defined mathematically. This would enable us to test out different
ideas on how to design our experiments. Table 1 shows the organizations of our
datasets.
 CapsNet vs CNN 5

(a) Arrows (b) Non-Arrows

Fig. 2. Samples from Group 1

(a) Triangles (b) Rectangles

Fig. 3. Samples from Group 2

Table 1. Organizations of our datasets

Dataset Description Subsets Number of images

Group 1 Contain images of Training Set 1 500
arrows and Training Set 2 1000
non-arrows
Training Set 3 2000
Testing Set 2000
Group 2 Contain images of Training Set 1 500
equilateral triangles Training Set 2 1000
and rectangles
Training Set 3 2000
Testing Set 2000

3.2 Convolutional Neural Network (CNN)

We implemented a CNN model using Tensorflow that has 3 convolutional layers
and 2 fully-connected layers. Max pooling layer was implemented after each
convolutional layer. Rectified Linear Unit (ReLU) was used as the activation
function on every layer except for the output layer. Dropouts were also applied
to prevent the model from overfitting.
As mentioned in the methodology section above, we carried out our exper-
iment by first training our CNN model on the dataset from Group 1 and once
the model was trained, we re-trained the weights of the fully-connected layers of
the model on the dataset from Group 2 while freezing the weights of the con-
volutional layers. After each training, the trained model was evaluated on the
testing sets from their respective groups.

3.3 Capsule Network with Dynamic Routing (CapsNet)

Our CapsNet model was also implemented using Tensorflow. We implemented
3 convolutional layers, 1 primary capsule layer and 1 digit capsule layer. The
6 U. Manogaran et al.

architecture of CapsNet is similar to the original paper [11] except that we added
an extra convolutional layer and we used 16-D capsules on primary capsule layer
and 32-D capsules in digit capsule layer. We used the activation function as
proposed in the paper. To prevent overfitting, a reconstruction network [11] was
used. There were no pooling layers used.
Similar to CNN, our experiment was carried out by first training our CapsNet
model on the dataset from Group 1 and once the model was trained, we re-trained
the weights of the primary capsule layer and digit capsule layer of the model on
the dataset from Group 2 while freezing the weights of the convolutional layers.
The trained model was evaluated on the testing sets from their respective groups
after each training.

4 Experimental Results and Discussion

The trainings and the evaluations of the models were performed on a workstation
running on Ubuntu 16.04 equipped with 16 GB RAM, Core i7-6700K processor,
and two NVIDIA GTX1080Ti GPUs. The models were trained using the training
subsets and were evaluated on their respective testing sets. The evaluation results
in terms of accuracy (acc), precision (prec), recall (rec) and F1-score (F1) for
both models are shown in Table 2 below.

Table 2. (a) Evaluation Results for CapsNet. (b) Evaluation Results for CNN

(a)
Subset 1 Subset 2 Subset 3
(%) Acc Prec Rec F1 Acc Prec Rec F1 Acc Prec Rec F1
Group 1
Triangles vs 88.4 88.7 87.2 87.9 91 92.5 89 90.7 90.4 92.8 87.6 90.1
Rectangles
Group 2
Arrows vs 67.6 70.6 59.7 64.7 77.8 86.6 62.1 72.3 80.4 83.9 75.1 79.3
Non-Arrows

(b)
Subset 1 Subset 2 Subset 3
(%) Acc Prec Rec F1 Acc Prec Rec F1 Acc Prec Rec F1
Group 1
Triangles vs 98.5 99.4 97.1 98.2 99.3 99.5 98.9 99.2 99.6 99.8 99.2 99.6
Rectangles
Group 2
Arrows vs 92.6 95 87.2 90.9 95.8 95.2 95.8 95.5 96.6 96.8 92.5 94.6
Non-Arrows
 CapsNet vs CNN 7

All the images were shuffled in their respective sets and normalized before
they were used for training and evaluation purposes. From Table 2(a), it is evi-
dent that CapsNet is able to determine whether a given image contains an arrow
or non-arrow by computing the spatial relationship between the learned features.
It can also be seen in Table 2 (b) that CNN has achieved near-perfect accuracies.
This is due to the fact that the generated datasets do not contain any real-world
noise.
We expected the CNN model to perform worse than CapsNet based on the
speculation stated earlier but it can be seen from the results that CNN actually
performed better than CapsNet. This might be due to the dataset being too
simple hence not requiring a deeper CNN model.
The use of pooling layers in between the convolutional layers should cause
the loss of spatial information of the features in a CNN. Hence, it might be
the case where our model is not deep enough. We expected our CNN model to
perform poorly to at least some degree due to the use of 3 pooling layers but
based on the results this is not the case. We chose a CNN model with only 3
pooling layers due to the simplicity of the datasets. From the results, it is evident
that the problem of retaining the spatial relationship between features is not a
serious issue for a relatively shallow model such as a model with only 3 pooling
layers. However, it is questionable whether a deeper CNN model would perform
well on a more complex dataset or not.
In our experiment, the objects in the images are formed by composing simple
features. There is only one equilateral triangle and one rectangle in every image.
Given the success of CNN, identifying such generated simple objects without
real-world noise is rather a trivial task for CNN. This could be another reason for
the high accuracy that CNN models have achieved in this experiment despite the
use of pooling layers. Our implementations are publicly available in this github
link.1

5 Conclusions and Future Work

In this work, we have designed an experiment to test the speculation that Cap-
sNet is able to retain the spatial relationship between features better than CNN.
In order to carry out the experiment, we have generated our own datasets.
From our results, both the shallow CNN and CapsNet models have shown
the capability to retain the spatial relationship between features. However, the
speculation that CapsNet is able to retain spatial relationship between features
better than CNN does not seem to be true for shallow models on simple datasets.
It is still uncertain whether this speculation is true for deeper models on more
complex datasets and on noisy datasets.
Considering the fact that CNN has been developed extensively since its inven-
tion in 1989 [12], it is possible that our experiment was too simple for CNN.
CapsNet on the other hand, is still at a rudimentary stage and the fact that its

1
https://github.com/MMU-VisionLab/CapsNet-vs-CNN.
8 U. Manogaran et al.

performance level is close to CNN in this experiment means that CapsNet has
great potential.
Future research in this area should consider the usage of more complex fea-
tures to represent the objects in the datasets and deeper models in order to
further understand the capabilities and limitations of these models. Gaining
deeper insights on these models in the retention of spatial relationship between
features will guide future developments in a better way.

Acknowledgment. The authors are grateful to the Ministry of Higher Education,

Malaysia and Multimedia University for the financial support provided by the Funda-
mental Research Grant Scheme (MMUE/150030) and MMU Internal Grant Scheme
(MMUI/170110).

References
1. Krizhevsky, A., Sutskever, I., Hinton, G.E.: ImageNet classification with deep con-
volutional neural networks. In: Advances in Neural Information Processing Sys-
tems, pp. 1097–1105 (2012)
2. Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Rabinovich,
A.: Going deeper with convolutions. In: Proceedings of the IEEE Conference on
Computer Vision and Pattern Recognition, pp. 1–9 (2015)
3. Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale
image recognition. arXiv preprint arXiv:1409.1556 (2014)
4. Redmon, J., Divvala, S., Girshick, R., Farhadi, A.: You only look once: unified,
real-time object detection. In: Proceedings of the IEEE Conference on Computer
Vision and Pattern Recognition, pp. 779–788 (2016)
5. Girshick, R., Donahue, J., Darrell, T., Malik, J.: Rich feature hierarchies for accu-
rate object detection and semantic segmentation. In: Proceedings of the IEEE
Conference on Computer Vision and Pattern Recognition, pp. 580–587 (2014)
6. Nair, P., Doshi, R., Keselj, S.: Pushing the limits of capsule networks. Technical
note (2018)
7. Algamdi, A.M., Sanchez, V., Li, C.T.: Learning temporal information from spatial
information using CapsNets for human action recognition. In: 2019 IEEE Interna-
tional Conference on Acoustics, Speech and Signal Processing (ICASSP), ICASSP
2019, pp. 3867–3871 (2019)
8. Xi, E., Bing, S., Jin, Y.: Capsule network performance on complex data. arXiv
preprint arXiv:1712.03480 (2017)
9. Xiang, C., Zhang, L., Tang, Y., Zou, W., Xu, C.: MS-CapsNet: a novel multi-scale
capsule network. IEEE Signal Process. Lett. 25(12), 1850–1854 (2018)
10. Chidester, B., Do, M.N., Ma, J.: Rotation equivariance and invariance in convolu-
tional neural networks. arXiv preprint arXiv:1805.12301 (2018)
11. Sabour, S., Frosst, N., Hinton, G.E.: Dynamic routing between capsules. In:
Advances in Neural Information Processing Systems, pp. 3856–3866 (2017)
12. LeCun, Y., Boser, B., Denker, J.S., Henderson, D., Howard, R.E., Hubbard, W.,
Jackel, L.D.: Backpropagation applied to handwritten zip code recognition. Neural
Comput. 1(4), 541–551 (1989)
13. Deng, J., Dong, W., Socher, R., Li, L.J., Li, K., Fei-Fei, L.: ImageNet: a large-
scale hierarchical image database. In: IEEE Conference on Computer Vision and
Pattern Recognition (2009)
Visit https://textbookfull.com
now to explore a rich
collection of eBooks, textbook
and enjoy exciting offers!
 CapsNet vs CNN 9

14. Palaz, D., Magimai-Doss, M., Collobert, R.: Analysis of CNN-based speech recog-
nition system using raw speech as input. In: Sixteenth Annual Conference of the
International Speech Communication Association (2015)
15. Zhang, C., Liu, W., Ma, H., Fu, H.: Siamese neural network based gait recognition
for human identification. In: IEEE International Conference on Acoustics, Speech
and Signal Processing (ICASSP), pp. 2832–2836 (2016)
16. Bojarski, M., Del Testa, D., Dworakowski, D., Firner, B., Flepp, B., Goyal,
P., Zhang, X.: End to end learning for self-driving cars. arXiv preprint
arXiv:1604.07316 (2016)
17. Tzelepi, M., Tefas, A.: Human crowd detection for drone flight safety using
convolutional neural networks. In: 25th European Signal Processing Conference
(EUSIPCO), pp. 743–747. IEEE (2017)
18. Milletari, F., Navab, N., Ahmadi, S.A.: V-net: fully convolutional neural networks
for volumetric medical image segmentation. In: IEEE Fourth International Con-
ference on 3D Vision (3DV), pp. 565–571 (2016)
19. Hinton, G.E., Krizhevsky, A., Wang, S.D.: Transforming auto-encoders. In: Pro-
ceedings of the 21th International Conference on Artificial Neural Networks-
Volume Part I, pp. 44–51 (2011)
20. LaLonde, R., Bagci, U.: Capsules for object segmentation. arXiv preprint
arXiv:1804.04241 (2018)
 Improved 2D Human Pose Tracking
Using Optical Flow Analysis

Aleksander Khelvas1(B) , Alexander Gilya-Zetinov1 , Egor Konyagin2 ,

Darya Demyanova1 , Pavel Sorokin1 , and Roman Khafizov1
1
Moscow Institute of Physics and Technologies, Dolgoprudnii, Russian Federation
khelvas.av@phystech.edu
2
Skolkovo Institute of Science and Technology,
Bolshoy Boulevard 30, bld. 1, Moscow, Russian Federation
http://www.mipt.ru

Abstract. In this paper, we propose a novel human body pose refine-

ment method that relies on an existing single-frame pose detector and
uses an optical flow algorithm in order to increase quality of output tra-
jectories. First, a pose estimation algorithm such as OpenPose is applied
and the error of keypoint position measurement is calculated. Then, the
velocity of each keypoint in frame coordinate space is estimated by an
optical flow algorithm, and results are merged through a Kalman filter.
The resulting trajectories for a set of experimental videos were calculated
and evaluated by metrics, which showed a positive impact of optical flow
velocity estimations. Our algorithm may be used as a preliminary step
to further joint trajectory processing, such as action recognition.

Keywords: Video processing · Human pose detection · Skeleton

motion

1 Introduction
Human motion tracking is an important application of machine vision algorithms
that could be used for many business purposes. The most popular tasks in the
digital world include distributed video surveillance system, solutions for digital
marketing, solutions for human tracking in an industrial environment.
This task can have different levels of details. The high-level approach is object
detection, when the position of human as a whole object is extracted and its
bounding box in 2D or 3D space is estimated.
A more interesting approach would be to detect a human pose in motion. This
task is more complicated because human pose has substantially more dimensions
compared to a bounding box.
Recent advances in deep learning have resulted in efficient single-frame pose
tracking algorithms, such as [6,14]. By applying them sequentially to a video
stream, a set of trajectories for joints may be obtained. However, since these

c Springer Nature Switzerland AG 2021

K. Arai et al. (Eds.): IntelliSys 2020, AISC 1251, pp. 10–22, 2021.
https://doi.org/10.1007/978-3-030-55187-2_2
 Improved 2D Human Pose Tracking Using Optical Flow Analysis 11

algorithms usually analyze input frames independently, the obtained trajectories

usually have various artifacts, such as discontinuities or missing points.
In the reported research, we solve a task of enhancing obtained joint tra-
jectories for multiple persons in a scene by leveraging the temporal information
using an optical flow algorithm.

2 Related Work

The task of retrieving pose dynamics for all persons in the video may be con-
sidered as a variant of multiple object tracking (MOT) task, where the con-
sidered objects are not persons but individual pose keypoints. There are two
major paradigms in the field of MOT - detection-based tracking and detection-
free tracking [11]. In the first case, machine vision algorithm capable of detect-
ing individual objects is applied to every frame separately and then individual
detections are linked into trajectories. The second approach has no detection
algorithm and instead relies on temporal changes in the video stream to detect
objects. With the development of efficient real-time object detection algorithms
in recent years, the detection-based approach has become dominant in the lit-
erature. However, independent analysis of video frames results in inevitable loss
of information conveyed by temporal changes in the video. This information
may be relevant to object detection and could help improve the tracker perfor-
mance. Various approaches were suggested to combine these individual frame
and temporal features.
For example, in [12] a novel approach to combine temporal and spatial fea-
tures was proposed by adding recurrent temporal component to a convolutional
neural network (CNN) designed to detect objects in a single frame. The outputs
of object detection network in sequential frames were fed into recurrent neural
network (RNN). The resulting architecture is then trained to predict the refined
tracking locations.
In [1] a tracker using prior information about possible person pose dynamics
is proposed. This information is modelled as a hierarchical Gaussian process
latent variable model, and allows to impose some temporal coherency in detected
articulations.
In [17] a method leveraging optical flow for a pose tracking is proposed. The
velocities obtained from flow data are used to generate expected coordinates of
a pose in next frame. Predicted coordinates are used later to form tracks by
greedy matching.
Our research is based on OpenPose as a body pose detector, proposed in
[3]. It is a real-time solution capable to detect a 2D pose of multiple people
in an image. It uses a non-parametric representation, which is referred to as
Part Affinity Fields (PAFs), to learn to associate body parts with individuals in
the image. This bottom-up system achieves high accuracy and real-time perfor-
mance, regardless of the number of people in the image.
12 A. Khelvas et al.

3 Definitions
Fist let us define several frames of reference (FoR) for our research, which are
shown in Fig. 1.

Fig. 1. Frames of references for 2D skeletons parameters calculation

U, V – this frame of reference is associated with virtual or real motionless

camera.
Ucf , Vcf – this frame of reference is associated with frames in video. If camera
is motionless, this FoR will be the same for all frames in video. This would be a
common case of video surveillance systems for security or marketing.
Upf k , Vpf k – this frame of reference is associated with object k, detected for
the frame f .
We will not use index f for video processing of motionless camera viewed
scenes.

4 Method

Our goal is to propose a novel algorithm for robust tracking of multiple person
poses in the video stream by leveraging both temporal and spatial features of the
data. To achieve this, we combine predictions done by a single-frame person/pose
detection algorithm (such as OpenPose and YOLO) with Optical Flow - based
estimations through a Kalman filter.
The complete algorithm is described below and shown in Fig. 2.
 Improved 2D Human Pose Tracking Using Optical Flow Analysis 13

Fig. 2. Full algorithm for 2D skeleton model calculation and filtration

1. Video preliminary processing step produces a set of frames with normalized

brightness/contrast and calculates Gf values.

2. Objects detection step provides a set of bounding boxes for each person,
detected by YOLO or some other object detection algorithm.

3. Pose detection ROI generation step provides a set of input frame regions for
further pose detection.

4. 2D pose estimation and person identification step computes a set of vectors

B f p = {uf p , v f p , uf p , v f p , . . . , uf p , v f p , }, where N = 25 is the number of joints
1 1 2 2 N N
for selected model of human body. (BODY-25 model provided by the OpenPose
solution)

5. Optical Flow calculation step applies an optical flow estimation algorithm to

the input frame, producing pixel velocity vectors for every joint position returned
by the pose detector.
Random documents with unrelated
content Scribd suggests to you:
lectures on the “Teacher’s Practical Philosophy”; the class in Kyoto
which entered for a course of twenty hours on the same subject
numbered rather more than eight hundred, and of these nearly
seven hundred and fifty received certificates for constancy in
attendance. At Nagasaki, Sendai, and other places, similar classes
obtained and kept an average attendance of from four hundred to six
hundred. At the close of each of these engagements, the class,
together with their foreign teacher, always had to be photographed.
It is well known to all travellers in Japan, and to all readers of books
on Japan, how much the Japanese, in their intercourse with each
other, insist upon a formal and elaborate politeness; and how careful
the better classes, and even the body of the common people, are to
practice this virtue, so esteemed by them, in all their intercourse with
foreigners. But it is far from being generally or sufficiently
recognised, how unfortunate and even positively shocking, the
disregard—not of their particular forms, but of all attempts at the
polite treatment of others, seems to them, as they are so constantly
forced to notice its prevalence among foreigners. That a fair degree
of genuineness attaches itself to these formal and conventional
observances, no one who knows the nation at all thoroughly can for
an instant entertain a doubt. Of course, on the other hand, neither
non-compliance nor the most exact compliance, mean the same
thing with us as with the Japanese. With them, not to treat a person
—even a coolie—politely, is positively to insult him. The foreigner
who should treat the native domestic servant, when the latter
approached on his knees and bowing his head constantly to the
floor, with an insult or a blow, might pay the penalty with his life. But
the old-fashioned politeness is being put to a difficult test by the
conditions of modern life, and by the changes of costume and of
customs which are being introduced from abroad. It may seem
strange that I speak of changes in costume as influencing the rules
for polite social intercourse. But, for example, the Japanese kimono
forms a fitting and convenient clothing for ladies who, on indoor
festal occasions, salute each other by hitching along the floor on
their knees, bowing the head as low as possible at frequent intervals.
It is decidedly not so fitting and convenient, however, where courtesy
while standing is demanded by politeness; or where it is desired to
dance with decency and elegance. On the other hand, the modern
gown, whether with or without train, is even less well adapted to the
practice of the requirements of the native social ceremonial.
According to the Japanese ideas, a proper respect for the teacher
requires that the pupil should receive and salute him, while standing.
This rule characterises the ceremonial adopted by audiences of all
sizes and as composed of different classes of hearers. In all the
lectures before audiences composed principally of teachers—since
they were, of course, for the time being regarded as the pupils of the
lecturer—the procedure was as follows: A select few, such as the
President of the Imperial or of the local Teacher’s Association, the
Mayor of the city, or his representative, and one or more members of
the Committee who had the affair in charge, were gathered some
time before the lecture-hour for tea-drinking in the reception room,
with the lecturer. At the appointed time—usually a little after, and
sometimes much after—this party of the select few proceeds to the
audience-room. On their entering the room, the entire audience rises
to its feet and remains standing until the speaker has mounted the
platform, bows have been interchanged with him, and he has sat
down. At the close of the address, the audience rises, bows are
again interchanged, and the “teacher,” unless some special
arrangement has been made and announced for him to remain for
further exercises, or to be introduced, leaves the hall first. The
audience is expected to remain standing until he has disappeared
through the door; it would be very impolite for them to begin sooner
to disperse. Indeed, I have never seen my friend, Baron T—, so
excited by anything else as he was on one occasion, when the
assembly of teachers began to move from their ranks, with the
appearance of breaking up, while I was only half-way between the
platform and the door.
It would be a great mistake, however, to infer from such passivity
and enduring patience in attention that Japanese audiences are
ready to accept with complaisance whatever any one may choose to
tell them for truth; or, indeed, to regard the ipse dixit of their
authorised instructors as of itself, a sufficient authority. On the
contrary, no small portion of the “Young Japan,” especially among
the student classes, is inclined to an extreme of bumptiousness.
Considering the circumstances of the present and the experiences of
the recent past, this is not strange; in view of the characteristics of
the race and its history in the more remote times, it is not unnatural.
Moreover, science, scholarship, and inventive talents, cannot be
subjected promptly in Japan to the same severe and decisive tests,
which are available to some extent in this country, but to a far greater
extent in most countries of Europe. But surely, in this country to-day
the difference between pretence or quackery and real merit or
unusual attainments, is not so well recognised, either by the people,
or by the press, or even by the executive officers of our educational
institutions, as to enable us to throw stones at the Japanese,—or for
that matter at any other civilised nation.
Perhaps there is no larger proportion of any Japanese audience,
who have perfect confidence in the superiority of their own views, or
in the originality and conclusiveness of their own trains of thinking, or
in their infallibility of judgment and loftiness of point of standing, than
would be the case with an audience similarly gathered and
constituted in America. I do not mean to say that Japanese student
audiences are lacking in docility or difficult to teach. On the contrary,
I think they are much more eager to hear about the last things in
science, politics, philosophy, and religion, than are the college and
university students in this country. And they certainly are on the
whole much more in deadly earnest in the matter of getting an
education. Something—probably much—of the old Samurai spirit still
lingers, which forbade the boy to rest or sleep until he had finished
his appointed task. I have had more than one of my own pupils tell
me how he had studied on through the night, applying wet bandages
to his head, or placing some sharp instrument so as to prick his
forehead, if, overcome by sleepiness, he nodded in his task.
This earnestness of demeanour, joined with the full confidence in an
ability to judge or even to discover for one’s self, undoubtedly makes
the audiences of students in Japan the more exacting. Besides, they
are prompt, severe, and even extreme,—oftentimes—in their
judgments concerning the ability and moral character of their
teachers. It is a by no means unusual occurrence for the students in
private or even in the Government institutions, to demand the
removal of some teacher, about whom they have made up their
minds that he is either incompetent as a scholar or unsafe and
misleading as a guide. I have repeatedly heard of “strikes” among
the students to enforce such a demand; but I have yet to hear of a
strike or a “call-off” in the interests of fewer hours or easier lessons.
Indeed, nine-tenths of the students in the Imperial University of
Tokyo are probably, in their ignorant enthusiasm to master quickly
the whole realm of learning, taking a much larger quotum of lecture-
hours than is for the good of sound scholarship, or than should be
permitted by the University.
As underlying or supporting or modifying all the other characteristic
features of the task attempted by the foreigner who expects to be
really successful in treating of serious themes with a Japanese
audience, is the high value placed on education by the nation at
large. At the period of first excitement over the action of the School
Board of San Francisco, in 1906, a Japanese friend of mine, a
professor in the Imperial University of Tokyo, who had spent some
fifteen years of his earlier life in this country, remarked to me with
extreme concern and sadness, that now his countrymen were
wounded by us at their most sensitive point. “Nothing else,” he
added, “do all our common people prize so much, for their children
and for themselves, as education.” In spite of its comparative
poverty, and of the feeling which—wisely or unwisely—it shares with
America and Europe, that the lion’s part of its resources must go to
the support of the army and navy, there is none of these nations
which is giving so much official attention to the education of all its
people as is Japan. As has already been pointed out in another
connection, the minister of Education takes rank with the other
members of the Ministry. The President of the Imperial Teacher’s
Association is a member of the House of Peers; he is a permanent
officer and his office is not a merely honorary position, is in no
respect a sinecure. As I know very well, his active administration
includes the care of the details, physical and intellectual, of the
various meetings of the Association. The case is as though some
Government official of high rank—for example like the late Senator
Hoar of Massachusetts—were to be the permanent president and
active manager of the general Teachers’ Association of the United
States. The Professors of the Imperial Universities have court rank,
in accordance with the length of the time and the distinction of their
services. Distinguished men of science and of literature are
appointed members of the Upper House or are decorated by the
Emperor, in recognition of their services to the country and of the
value of their presence, as men who may be reasonably supposed to
know what they are talking about, in the councils of the nation.
Diplomats, even of the lower ranks, must be educated in the
languages and history of the countries in which they are to be
stationed as members of the foreign service. The ability to read,
speak, and write English is required of all the graduates of the
Government Schools of Trade and Commerce. There is a larger
proportion of the children in the public schools than in any other
country, with the possible exception of Germany. The proportion of
illiterates to the entire population is much less than it is in this
country. And in spite of the meagreness of equipment, the
incompetence of much of the teaching force, the large amount of
crude experimenting, and the numerous and serious deficiencies,
which still afflict the system of public education in Japan, the
recognition of the absolute necessity and supreme value of
education in determining the conditions of national prosperity and
even of continued national existence, is intelligent, sincere, and
practically operative among all classes throughout Japan.
Now this esteem of the importance of instruction has a profound, if
not consciously recognised, influence on the attitude of all sorts of
Japanese audiences toward the person who is addressing them. He
is assumed to be telling them something which is true and which
they need to know. Talks by foreigners, that are merely for
entertainment or amusement are, in general, an insoluble puzzle to
the average Japanese audience. Of course this failure to appreciate
such efforts is in part due to the wide difference in the spirit and
structure of the two languages,—that of the speaker and that of his
hearers. But it is even more largely due to something which lies far
deeper. The Oriental story-teller and professional joker has his place
in the estimate of the educated and even of the common people. It is
side by side with the juggler or performer with marionettes. To
consider such a person as a teacher would be foreign to the
conception,—a scandalous profanation of a sacred term. On the
other hand, if one can—and this must come only after a considerable
period of testing—win and hold the claim in its highest meaning, he
may, as simply a teacher, wield an influence in Japan which is
comparable to that to be gained in the same way in no other civilised
land. For have not the greatly and permanently influential
benefactors of the race always been teachers? Were not Confucius,
and Sakya-Muni, and Jesus, all teachers? And in Japan itself, only a
few years ago, did not a certain man who had refused offers of
government positions deemed higher by most, in order that he might
remain a teacher of Japanese youth, when his work was ended,
have his coffin attended to its resting-place by ten thousand of his
fellow citizens of all classes, walking bareheaded through the rain?
Even now, when the appreciation of the importance and value of
wealth, for the individual and for the nation, is rising, and the
appreciation of the importance and value of intelligence and
character is, I fear, relatively declining, it is still possible for the truly
successful teacher to gain the esteem and influence over Japanese
audiences which are implied in the very word Sensei, or its
equivalent. And if the title is used with its full, old-fashioned
significance, it will have much the same meaning as the word
“Master” in the New Testament usage.
 “THE BEARING OF THE BOYS AND GIRLS IS SERIOUS,
RESPECTFUL AND AFFECTIONATE”
At the time of my last visit to Japan, in 1906 and 1907, the temper of
the entire nation was particularly and indeed uniquely interesting.
They had just been through a terrible struggle with what had, at the
beginning of the struggle, been quite generally regarded as an
invincible European power. They had been, indeed, uniformly
victorious; but at the cost of enormous treasure and of the
outpouring of the blood of the flower of their youth. The nation was
heavily burdened with debt; and its credit, in spite of the fact that the
financing of the war had been conducted with very unusual honesty,
frankness, and skill, was low for purposes of borrowing large
additional sums of money. The great body of the people, who did not
know what His Majesty, the Genro, and the most intimate circle of
advisers knew perfectly well, considered the nation humiliated and
defrauded by the unfavourable terms on which the Portsmouth
Treaty of Peace was concluded. As I can testify, there was an almost
complete absence of those manifestations of elation and headiness,
amounting to over-confidence and excessive self-conceit, which
prevailed so widely at the end of the Chino-Japanese war. On the
contrary, the great body of the people, especially outside of Tokyo
and the ports of Yokohama and Kobe, were in a thoughtful, serious,
and even anxious state of mind. This condition could not fail to make
itself felt upon the attitude of the audiences toward those who
addressed them, in correspondingly thoughtful and serious fashion,
on themes of education, morals, and religion. Even in the public
schools of the primary grade, the bearing of the boys and girls
toward their work is serious; and toward their teachers, respectful
and even affectionate.
Indeed, in the year after the war with Russia ended, the demand
everywhere in Japan was for the discussion of moral problems; and
of educational, economic, and political problems, as affected by
moral conditions and moral principles. The lectures to the teachers
which were most eagerly welcomed and which made by far the most
profound impression, spoke of the teacher’s function, equipment,
ideals, and relations to society and to the state, from the ethical point
of view. A course of lectures on the “Doctrine of the Virtues as
applied to Modern Business” was called for by the Government
Business Colleges. On my accepting an invitation to speak to the
boys in the Fisheries Institute, and asking for the topic which was
preferred for the address, the reply was given without hesitation: “Tell
them that they must be ‘good men,’ and how they may serve their
country better by becoming good men. Most of these boys come
from low-class families, whose morals are very bad, and they have
not been well brought up; but we wish them to become honest and
virtuous men.”
Nor was this interest, amounting in many cases to anxiety, about the
moral condition and future moral welfare of the nation, confined to
educational circles. The Eleventh of February is a national holiday in
Japan, corresponding more exactly than any other of its national
holidays to our Fourth of July. This is the traditional date of the
founding of the Empire. On this date, in 1889, the “Constitution was
promulgated by the Emperor in person, with solemn and gorgeous
ceremony, in the throne-room of the Imperial Palace,—and its
proclamation was followed by national rejoicings and festivities.” In
1907 the Asahi Shimbun, a leading paper of the large commercial
city of Osaka, undertook to commemorate the day by a public
meeting in the Hall of the Assembly of the District, and by a banquet
in the neighbouring hotel. In the afternoon I addressed an audience
of more than twelve hundred on the “Conditions of National
Prosperity,” dwelling chiefly on those conditions which are
dominatingly moral and religious in character. The banquet in the
evening was attended by about one hundred and fifty persons, and
was fairly representative of all the most influential citizens of various
classes. After the customary exchange of complimentary addresses,
opportunity was given for others to speak. A venerable gentleman,
one of the most distinguished physicians of the city, was the first to
rise. With great seriousness he made, in substance, the following
comments on the exercises of the afternoon,—which were, however,
not interpreted for me until a full fortnight later. He had been much
impressed by what the speaker had said at the afternoon meeting
about the dependence of national prosperity upon the nation’s
morality; but he had asked himself: “Why are such things said to us?
We are not ‘rice merchants’ (a term of opprobrium); we are the
leading and most respectable citizens of Osaka.” He had, however,
at once reminded himself that this is precisely what their own great
teacher, Confucius, taught them centuries ago. And then he had
asked himself: “Why do the ancient Oriental teacher, and the modern
teacher, both teach the same thing,—namely, that nations can have
genuine and lasting prosperity only on condition that they continue to
pattern themselves after the eternal principles of righteousness?” His
answer was: “They tell us this, because it is so.” And “surely,” he
added with much impressiveness, “it is time that we were all
governing our actions in accordance with so important a truth.”
After the aged speaker had taken his seat again, a much younger
man, the Vice-Mayor of the city, arose; and beginning by expressing
his hearty agreement with the sentiments of the last speaker, he
proceeded to emphasise the truth with passionate fervour, and
wound up his address by saying: “There are enough of us, one
hundred and fifty leading citizens of Osaka, seated around this table
here to-night, to change the whole moral condition of the city, and to
redeem it from its deservedly bad reputation, if only we truly and
fixedly will to have it so.”
Several months later I had another similar experience, which I
mention here, because it illustrates so well the extraordinary interest
in moral issues which characterised the disposition of the nation at
the close of the Russo-Japanese war, and which made itself felt in
so powerful a way upon all the audiences which I addressed during
the year of my stay. Toward the close of the course of lectures and
addresses at Sendai, I was invited to visit the barracks where
twenty-five or thirty thousand of the recruits for the Japanese army
are regularly undergoing their preparation for service. After I had
been shown about the entire establishment by an escort of under-
officers, the General in command, a distinguished veteran of the
Russo-Japanese war, called me into his private office. There, he first
of all assured me that he had followed the accounts of the lectures
and addresses as they had been published in the various papers,
and then thanked me for what had been done in general for the good
of his country; but, more particularly, for the assistance rendered to
him personally in his work of training the young men for the
Japanese army. Upon surprise being expressed as to how such a
thing could be, the General began to explain his statement as
follows: His great difficulty was not in teaching the manual of arms or
the proper way to manœuvre upon the field of battle. His great
difficulty was in giving these recruits the necessary “spiritual” training
(I use his word, and explain it to mean,—The moral spirit which
animates the upright and knightly soldier, the spirit which, in the
Japanese language, is called “Bushidō”). At this I again expressed
surprise and a wish for further light upon his kindly remark. He then
went on to say that since the Government had reduced the term of
required service from three years to two, the time was more than
ever all too short to inculcate and enforce the right moral spirit on
youths, many of whom came from homes in which this spirit by no
means prevailed. But a profound moral impression had been made
upon the teachers in the public schools all over the land; the
teachers would take these moral teachings and impress them upon
the pupils under their charge; and “these are the boys that will later
come to me.” When my thoughts turned homeward—as, of course,
they were bound promptly to do,—they awakened a strange mixture
of feelings of amusement and of concern; of the former, when the
effort was made to imagine any remotely similar conversation
occurring with a General or a recruiting officer there; and of concern,
at the obvious decline of the spirit of patriotism in the United States,
as evinced by the almost purely mercenary way in which all
branches of the public service have come to be regarded by the
body of the people. That it is difficult to keep the ranks of our small
standing army filled by offers of big pay, much leisure, and
opportunities for foreign travel, is significant, not so much because of
a growing and, perhaps, reasonable distaste for a military career, as
of a prevalent conviction that the nation is bound to serve individual
and class interests rather than the individual and the class to serve
the interests of the nation.
The most thoughtful leaders of Japan are at present exceedingly
fearful that those more serious and self-reliant traits, which I have
chiefly selected to characterise, may succumb to the incoming flood
of commercial avarice, and of the love of comfort and luxury. And
they have grave reasons to be afraid. But I leave on record my
testimony to the truth that immediately after the close of the great
war with Russia, the nation of Japan was not only willing to hear, but
even coveted to hear, how it might prepare itself by intelligent
adherence to sound moral principles—in education, in business, in
the army and navy,—for an era of genuine and lasting prosperity, at
peace with the rest of the world.
 CHAPTER VI
GARDENS AND GARDEN PARTIES

To understand thoroughly and appreciate justly the theory and

history of the art of landscape and other gardening in Japan would
require the study of a life-time. It is doubtful if any foreigner could
accomplish this task even at the expense of so great devotion;—so
subtle and in some respects bizarre and whimsical is the philosophy
of nature implied in the tenets of some of the various schools. The
native experts, too, take the same delight in minute distinctions, and
in the arguments urged in support of them, in the field of æsthetics,
which characterises the speculations of Japanese Buddhism in the
field of religion. And, indeed, in Japan, as everywhere else in the
world, the philosophy of nature and the philosophy of art are both
closely related to ideas and sentiments of at least a genuine quasi-
religious character. I shall therefore make no attempt to discourse on
so abstruse and difficult matters.
It may safely be said, however, that the art of landscape-gardening,
as it has developed in Japan, has the general features which are
common to all forms of Japanese art. According to Baron Kuki,
official custodian of the Emperor’s art treasures, these are, chiefly,
the following three: “The first is mildness and pure simplicity.
Colouring is for the most part sober and plain, and very seldom
gorgeous. Japanese art prefers moderation and genial ease to
excessive grandeur; sobriety and chastity to profundity, intensity, and
vulgarity. Even such horror-inspiring subjects as the pictures of hell
are not thrilling in effect. The statue of Buddha at Nara is grand, but
it is only the highwater mark of Continental influence, and does not
represent the pure Japanese disposition.
“The second characteristic of Japanese art is its exquisite lightness
or delicacy. This is due to the joyful frame of the people’s mind, and
to the wonderful dexterity of their hands. There is no artistic product
which is not marked by charming workmanship.
“The third feature is its idealism in representation. Japanese art is
not realistic, it does not aim at photographic accuracy; but by the free
and bold exercise of imagination, it tries to abstract the essential
aspect of objects, and to give expression to the artist’s sentiments by
its portrayal. It is for this reason that form is comparatively little
regarded, while idea is considered all-important; that it is weak in
realistic delineation and strong in decorative design. These three
characteristics underlie all Japanese art, and distinguish it from the
art of other Oriental nations.”
So far as I am able to recognise these characteristics as present in
the gardens which I have seen, they have resulted in certain marked
excellencies and in certain scarcely less marked defects. These
gardens are to a degree realistic, in that they try to present a picture
of all the principal features of nature,—oftentimes, and indeed
generally, within a small and seemingly inadequate amount of space.
Miniature mountains, rivers, lakes, water-falls, forest, and stretches
of sea-coast, may be comprised within the grounds of a gentleman’s
ordinary estate, or even within the few square-feet of the humbler
citizen’s back-yard. And it is uniformly the back-yard in Japan, where
the grounds for “plaisance” are situated. Even a platter or other dish
may be made the receptacle for a garden which shall essay to hold
up to view a picture of those complex artistic achievements that are
accomplished by nature on a so much larger scale.
“IT IS NATURE COMBED AND TRIMMED”
It is not “pure nature,” or nature untamed and wild, which the
Japanese art of gardening aims simply to reproduce and to
represent. It is nature excessively combed and trimmed; or—to
present the thought in more carefully chosen æsthetical language—
nature as she would be if arranged and arrayed according to the
most precisely developed ideals of the human artist. Every tree and
shrub must, then, be cultivated and pruned with attention to the
details of each stem and twig; even the decayed or superfluous
leaves, or the needles and cones from the pine trees, require to be
picked away. On visiting the famous garden of Count Okuma, in the
late Autumn, we found two-score and more of persons, working
under expert direction in this way. The same pains is taken in holding
up to view the work of nature in marring her own products or in
removing them utterly, in order to make room for the fresh creations
of her bounteous life. Worn rocks, worm-eaten woods, bare trunks,
broken stumps, and all the other results of the ceaseless forces that
minister decay and death, have an important place in the Japanese
art of gardening. But the worn stones must be carefully placed and
kept scrupulously clean; the worm-eaten woods must be selected
with due regard to the fantastic patterns which have been worked
upon them; the bare trunks and broken stumps need to have their
shapes defined by the back-ground of foliage or of open sky; for—to
quote again from Baron Kuki—the effort is to “abstract the essential
aspect of objects, and to give expression to the artist’s sentiments by
their portrayal.”
Only a little reflection is necessary in order to make it evident that for
the æsthetical appreciation of the Japanese garden, in its most
purely native form, whether as originally imported from China or as
developed on native soil, a sympathetic share in this characteristic,
sentimental attitude toward nature is absolutely indispensable. In
viewing the best examples, where the scale is fairly generous and
the artistic theories in control have not been too individualistic or
fantastic, such sympathy is not difficult for one of cultivated
æsthetical taste; although the Japanese art may still make the
impression of being something unusual and foreign. Where,
however, such sympathy is wanting, and in the cases of multitudes
of inferior examples, no amount of this feeling—or, at least, no
reasonable amount—can easily prevent an unfavorable judgment,
on account of the impression of artificiality, pettiness, and excessive
devotion to details, without a corresponding largeness of spirit. But
when one recognises the amount of innocent enjoyment, and of a
valuable sort of æsthetical education and refinement—for the
Japanese garden is seldom or never vulgar,—which comes to the
homes of the lowly in this way, one’s criticism is either totally
disarmed or greatly modified in its points of view. Indeed, there is no
nation in Europe or America to whom the Japanese may not give
valuable lessons in the art of the quiet, soothing, and refining
enjoyment of nature and of out-of-door recreations, to the discredit
and relative neglect of those coarser and more exhausting ways of
enjoying themselves which these other nations prefer. Moon-viewing,
cherry-blossom viewing, and mushroom-gathering parties may seem
to us lacking in “strong” inducements, as modes of pleasure-seeking;
but the men and women who have made an art of cultivating them—
and these have been among the greatest in the history of Japan—
certainly can no longer be considered as a nation of dilettanti or of
weaklings.
It is a not uncommon impression, even on the part of those who
have visited the country, that Japan is a “land unrivalled in the
beauty and abundance of its flowers,—a belief that nature has
lavished her floral gifts with special favour upon these sunny islands
of the Far East.” But as Mr. Conder points out in his admirable book
upon “The Floral Art of Japan,” in the sense of “profusion in wild
floral plants, it must be admitted that certain Western countries
possess attractions which Japanese scenery can scarcely boast.”
And although, as he goes on to say, “the comparative scarcity of
groups of wild flowering plants, as a colour feature to the landscape,
is to some extent made up for by the blossoming trees,” the peculiar
characteristics and values of the Japanese art of gardening have not
been so much derived from the nature that is without as from the
nature that, centuries ago, lay slumbering within the spirit of the race.
It must also be remembered that, just as the Japanese floral art does
not confine itself to the æsthetical treatment of “flowers,” in our
narrower use of the word, but, the rather, includes all flora in the
botanical meaning of the term, so the art of gardening in Japan aims
to take account of all forms of material and of situations and even of
remote suggestions, which fall within the limits of man’s artistic
control. “The secret, then, of Japan’s floral fame and floral
enchantment lies rather in the care that her people bestow upon
Nature’s simpler gifts than in any transcendent wealth of production.”
“Flower-viewing excursions, together with such pastimes as Shell-
gathering, Mushroom-picking, and Moon-viewing, form the favourite
occupations of the holiday seeker throughout the year. By a pretty
fancy, even the snow-clad landscape is regarded as Winter’s floral
display, and Snow-viewing is included as one of the flower festivals
of the year. The Chinese calendar, used formerly by the Japanese,
fitted in admirably with the poetical succession of flowers. Haru, the
Japanese Spring, opened with the New Year, which commenced
about February, and was heralded by the appearance of the plum
blossoms.”
Floral art in Japan, therefore, makes extensive and effective use of
flowerless trees, as well as of flowers, and flowering shrubs and
flowering trees. Among such flowerless trees, the most important is
the pine; and this hardy evergreen is found almost everywhere in the
mountain and coast scenery of the country, and in all the gardens, as
well as in a large proportion of the floral arrangements designed for
in-door enjoyment. In its natural growth and struggle against the
violent winds, it is habitually so quaintly distorted that the miniature
representations in the smallest gardens and in tiny pots, are scarcely
at all exaggerated. Then follow, in order of preference, the bamboo,
the willow, and other flowerless trees.
Inside-floral arrangements should have regard to the character and
uses of the room in which they are placed, to the season of the year,
to the nature of the festival or other ceremonial occasion which they
may chance to celebrate, to the other art-objects and the furniture of
the same and adjoining rooms, and to the scenery of the garden and
the remoter landscape upon which the room opens. To quote again
from Mr. Conder: “Some writers go so far as to say that the floral
design in a chamber should have a contrast in style with that of the
adjoining garden. This fancy is better appreciated if it be
remembered that during a great part of the year the outer walls of
the Japanese house, which consist almost entirely of paper slides,
are thrown completely open. If there be a landscape garden
adjoining, consisting of lakes and hills,” (and as we have already
seen these objects may exist in exceedingly miniature form) “the
floral arrangement in the rooms should by preference partake of a
moorland character; but if the garden be level and waterless, then
water plants or mountain trees should be selected for the flower
decorations of the chamber interior.”
I have already said that a great deal of philosophy—originally
derived from China—together with not a few traditional superstitions,
underlies the art of floral arrangement and the allied art of gardening,
in Japan. But, what is more important in its influence upon the life of
the people, is this: The expression and cultivation of virtue, and of
the religious spirit,—of self-denial, gentleness, and the forgetfulness
of cares—are both theoretically and in practice realisable and
actually realised through this form of art.
Without retracting my previous disclaimer of the intention to venture
into the field of philosophy in its relation to the Japanese art of
landscape and other forms of gardening, I will make this final
quotation from Mr. Conder’s treatise on the subject of floral
arrangement, in one of the few passages where he extends his
observations to the wider fields of the art of gardening. He has been
speaking of the applications made of the male and female principles,
so often referred to in Confucian philosophy, to contrasts of forms,
surfaces, and colours, in the composition of floral material. “It has
ever been a favourite fancy of the Japanese to apply distinctions of
sex to inanimate nature. In natural scenery, and landscape-
gardening, it is customary to discriminate between male and female
cascades, male and female plants and trees, male and female rocks
and stones. The distinction is not one so much of individual and
separate quality as of forms placed in combination or contrast, and
regarded as male or female in respect of one another. Thus the main
torrent of a water-fall is considered masculine, and the lower fall in
proximity feminine. In like manner, rocks used in gardening have no
distinguishing sex, unless they are placed in pairs or groups. In the
case of two stones of different character placed side by side, the one
of bolder and more vigorous shape will be called the male, and the
other the female stone. Curious as such fancies may seem, they are
of considerable value when applied in the arts of design, their
observance helping to produce that harmony of well-balanced
contrasts which should pervade all artistic composition.”
Another striking illustration of the influence of quasi-moral and
religious sentiment over this form of art is to be seen in the use
made of the lotus in the landscape gardening of Japan. “The lotus is
closely connected with the Buddhist religion, and is, therefore,
associated in the minds of the people with spirit-land. The lakes of
the temple grounds, especially those dedicated to the water goddess
Benten, are frequently planted with lotuses.... Wherever undisturbed
pools and channels of muddy water exist, the lotus is to be found,
and even the ditches beside the railway connecting Tokyo with the
port of Yokohama are rendered gay in the Summer by the lotus
flowers in bloom. As the peony is said to be the national flower of
China, so the lotus is regarded as the national flower of India, the
source and centre of Buddhism. It is therefore considered out of
place as a decoration for occasions of festivity and rejoicing, but is
constantly used for obsequies and other sacred ceremonies. The
lotus serves as suitable theme for religious contemplation” (and
according to the psychologically true thought of the Japanese, the
most fit and profitable place for such mental exercises is in the open
air, and under the sane and soothing and uplifting influences of
nature) “and is the favourite flower of monastic and temple retreats;
the best displays are to be seen in the lakes of the old temple groves
of Kyoto and other cities. Growing out of the muddiest and most
stagnant water, its leaves and flowers are always fresh and clean;
although it is particularly sensitive and quickly withers if brought in
contact with any of the fertilisers by which other plants are
nourished. This purity which the lotus maintains amid surrounding
filth is mentioned as one reason for associating this plant with the
religious life. A well-known book of Buddhist precepts contains this
text:—‘If thou be born in the poor man’s hovel, but hast wisdom, then
art thou like the lotus flower growing out of the mud.’”
The most beautiful and perfect of the gardens of Japan, in the old-
fashioned Chino-Japanese style, which I have ever seen, are the
Kōrakuen in Tokyo, and the Katsura-no-Rikyū in Kyoto. The former
was originally the garden of the Prince of Mito, the site of whose
mansion is now occupied by the Koishikawa Arsenal. Thus the quiet
beauty of the art of the “Old Japan” is brought into contrast with the
preparations for displaying its strength of the “New Japan.” But the
garden remains intact; and it is justly pronounced “the finest
specimen of the Japanese landscape gardener’s art to be seen in
the capital.” It is not seen, however, by most visitors to Japan, both
because they do not take an interest in, or know where to look for
the best things, and also because a special order is necessary to
gain admittance to it. The very name is an embodiment of the finest
philosophical sentiment with regard to the relations in which the
leisurely enjoyment of nature stands to the sterner duties of a
devoted human life. It is derived from three Chinese words:—Kō
(“afterward”) Ra (“pleasure”), and Kuen (“garden”). It is, therefore, an
“afterward-pleasure-garden”;—the thought being that the wise man
has his anxieties earlier than others, is beforehand, so to say, in
thoughtful care; but his pleasures come later.
The original plan of Kōrakuen was to reproduce many of the scenes
of the country with which the literati were familiar—at least, by their
names. And Prince Mito had it laid out as a place in which to enjoy a
calm old age after a life of labour. One of its miniature lakes is copied
from a celebrated lake in China. A temple on a wooded hill is a
replica of a famous temple in Kyoto. Again, a bridge and zigzag path
lead to a shrine famous in Chinese history; and then we come to an
arched stone bridge and another shrine which has an octagonal
shape in allusion to the Eight Diagrams of the Chinese system of
divination. Everywhere there are magnificent trees, which were
selected so as to have some one species at the heighth of its beauty
at each season of the year; thus there are cherry-trees for the
Spring, maples for the Autumn, and plum-trees for the Winter.
An attendant who was to serve as an escort was already in waiting,
when we and the Japanese lady whom we accompanied arrived at
the gate; and somewhat later General Nishimura, the Government