![1 Introduction
Skin cancer incidence is increasing at 3.1% per year [1]. Skin cancer spread over
the body with the help of lymphatic and blood vessels. Thus, early detection of skin
cancer is very important for proper diagnosis of the disease.
Melanoma and non-melanoma are two major categories of skin cancers.
Malignant melanoma is of several subtypes. Basal cell carcinoma and squamous
cell carcinomas are two main types of non-melanoma skin cancers.
Each type of skin cancer is different from the other skin cancers in certain
characteristics.
In clinical detection of skin cancer diagnosis, dermatologists use a visual
inspection. Clinical diagnostic performance is very poor in comparison with der-
moscopy and automatic diagnosis. Dermoscopy is a noninvasive diagnostic tech-
nique. It uses clinical dermatology and dermatopathology in combination to inspect
the morphological features which is not possible in clinical detection. Dermoscopy
increases the performance of diagnosis with 10–30% compared to unaided eye [2].
Differentiation of skin cancer images needs much more experience with der-
moscopy technique. Less-experienced clinicians use ABCD-E rule [3] to improve
the diagnostic performance [3].
Automatic image processing of skin cancer gives better results by providing the
exact information about lesion, which can be useful for the clinician to detect and
classify skin cancer. It is also used as a stand-alone early warning tool. Effective
implementation of this automatic technique may give reduced deaths with benefits
both to the patients and to the healthcare system. Working of automatic skin cancer
detection has three main stages: (1) segmentation of lesion; (2) feature extraction
and feature selection; and (3) lesion classification.
Segmentation is an important process in image processing applications and
computer vision because doctors are always interested in meaningful regions of the
dermoscopy image. Segmentation divides an image into a number of separate
regions. Pixels, in each region, have high similarity such as color, intensity, and
texture. Many researchers use only gray level for image segmentation [31]. But, in
our proposed system, we use color information of the image for lesion
segmentation.
In general, we covert the color image in gray-level image; therefore, color
information does not used. There is a wide variety of segmentation methods used in
dermoscopy images [4]. Recent advancements include thresholding [5, 6], k-means
clustering [7], fuzzy c-means clustering [8, 9], density-based clustering [10],
mean-shift clustering [11], gradient vector flow snakes [12–14], color quantization
followed by spatial segmentation [15], statistical region merging [16], watershed
transformation [17], dynamic programming [18, 19], and supervised learning [20,
21]. Clustering is an unsupervised learning technique, where one can give the
number of clusters in advance to classify pixels [22]. A similarity measure is
defined between pixels, and similar pixels are then grouped into a cluster. We use
k-means clustering for segmentation of color images.
2 M. Ritesh and S. Ashwani](https://image.slidesharecdn.com/26279-250427032823-aa6dc0e6/85/Sensors-and-Image-Processing-Proceedings-of-CSI-2015-1st-Edition-Shabana-Urooj-19-320.jpg)
![It is very hard to differentiate skin cancer visually. Identification and extraction
of most effective features from cancer-affected lesion is very important. Each class
of skin cancer has some different features than others. We use these different
features for classification. Feature extraction extracts useful features or properties
from original dataset of an image. These extracted features easily classify the
classes of skin cancer.
Color features are mainly statistical parameters. These are calculated from inter-
and intra-channel of an image, like average value and standard deviation of the
RGB [23–27] or HSV color channels [28]. Here, we use “local binary pat-
terns + color percentiles,” “integrative co-occurrence matrices,” “gray-level
co-occurrence matrices + color percentiles,” “Gabor features + chromatic fea-
tures,” “Gabor features,” “opponent color LBP,” “color ranklets” [29]. These
methods are based on texture and disjoint color analysis. Textural features are
extracted from images by converting into gray level, and color features are com-
puted with the help of three color component of an image. Textural and color
features are concatenated into the same feature vector to improve the classification
accuracy.
The main aim of feature selection is to select the maximum number of features to
achieve high performance in cancer classification [30]. Feature selection is
important when anyone works on gray-level features. In our proposed system, there
is no need of feature selection algorithms.
Classification phase of the diagnostic system is the one in charge of making the
inferences about the extracted information in the previous phases in order to be able
to produce a diagnostic about the input image. In our experiment, we have used four
well-established classifiers: support vector classifier (SVC), nearest neighborhood
(NN), linear classifier, and nearest mean classifier (NMC).
The rest of the paper is organized as follows: Sect. 2 briefly reviews segmen-
tation, feature extraction techniques, and classification methods which are used in
proposed framework. Section 3 reports extensive experimental results, and Sect. 4
concludes the paper.
2 Proposed Framework
Proposed framework is a compiled abstraction of digital image classification.
Figure 1 describes the steps in an image classification.
In this paper, we propose this framework with k-means clustering segmentation
and color texture feature extraction techniques which is a new approach to classify
the skin cancer images. For validation purpose, twofold cross-validation is used.
A Comparative Study of Various Color Texture Features for Skin … 3](https://image.slidesharecdn.com/26279-250427032823-aa6dc0e6/85/Sensors-and-Image-Processing-Proceedings-of-CSI-2015-1st-Edition-Shabana-Urooj-20-320.jpg)
![2.1 Segmentation
Segmentation is a process to partitioning an image into disjoint regions that are
homogeneous with respect to a chosen property such as luminance, color, and
texture. The aim of segmentation is to change the representation of an image into
something that is more meaningful and easier to analyze [30].
K-means Clustering Segmentation: K-means clustering method is one of the
simplest unsupervised learning methods and this method is nondeterministic,
numerical, and iterative. In this experiment, color images are used as input; hence,
Fig. 1 Proposed method
4 M. Ritesh and S. Ashwani](https://image.slidesharecdn.com/26279-250427032823-aa6dc0e6/85/Sensors-and-Image-Processing-Proceedings-of-CSI-2015-1st-Edition-Shabana-Urooj-21-320.jpg)
![this technique is used for segmentation. K-means clustering is partitioning method.
This method groups objects in the way that within-group variance is minimized. If
within-group variance is minimized, then it gives high featured segmented image.
The working of this method is as follows [31, 32]:
1. Select K pixels for centers randomly. Initially, center represents group centroids.
2. For grouping the sample data, calculate histogram bin value distance between
pixels and selected centroids and assign the group on the basis of nearest
distance.
3. Calculate the histogram bin value for new group to find the new position of
centroids.
4. If the value of centroids changes, then repeat steps 2 and 3 (Fig. 2).
2.2 Feature Extraction
It is important to identify the most effective features to extract from skin cancer
lesions and to find the most effective pattern-classification criteria and algorithms
for differentiating those lesions.
Local Binary Patterns + Color Percentiles: The LBP feature vector is created in
the following way:
1. Divide the window size into cells in the form of n*
n matrix (i.e., 16*
16).
2. If the center pixel value (Threshold Value) is less than the selected pixel, then
write 1, otherwise write 0. For rotational invariant features, move in one
direction either clockwise or anticlockwise. This step gives 8-digit binary
number. Convert it into decimal number for further processing.
3. Compute the histogram of the frequency of each “number” occurring over the
given cell.
4. Apply normalization on histograms.
5. Concatenate the histograms of all divided cells.
The combination of LBP + CP presented by Niskanen et al. [33]. In this method,
we calculate local binary patterns in an grayscale image. For better results, we
reduce the calculated features into rotationally invariant features. Rotational
invariance is necessary because when image is rotated, the gray values also rotated
in a circular form where origin is same. Then, calculation of feature vector changes
Fig. 2 Input image and
segmented image
A Comparative Study of Various Color Texture Features for Skin … 5](https://image.slidesharecdn.com/26279-250427032823-aa6dc0e6/85/Sensors-and-Image-Processing-Proceedings-of-CSI-2015-1st-Edition-Shabana-Urooj-22-320.jpg)
![with the rotation. We apply discrete Fourier transformation function (DFT) to make
features rotationally invariant. There are always 36 rotationally invariant features in
a grayscale image. Now, we calculate the color percentiles of first, second, and third
quartiles (three points that divide the each channel into four parts) of three channels.
Total features from color percentiles are nine. When we concatenate features of
these two methods, the final resulting feature vector dimension is 36.
Integrative Co-occurrence Matrices: Integrative co-occurrence matrix [34–36] is
a new approach for calculating the inter-channel and intra-channel features. In
inter-channel feature calculation, we calculate the co-occurrence matrix features
from each of three channels, and in intra-channel, we calculate the features from
combination of color channels ((r,g), (r,b), (g,b)). In our experiment, we use five
co-occurrence features: energy, contrast, correlation, homogeneity, and inertia.
Final resulting feature vector dimension is 30.
Co-occurrence Matrices + Color Percentiles: The combination of co-occurrence
matrices and color percentiles is presented by Niskanen et al. [33]. This method is first
used for applications in wood inspection. Out of 14 features, we computed five
important co-occurrence features for classification. Color percentile is calculated as
discussed in LBPnCP method. So, the dimension offinal resulting feature vector is 14.
Gabor Features: Gabor features are used to extract scale and orientation infor-
mation from different channels. We calculate Gabor features for inter-channel of an
image with above-mentioned configuration. One channel gives 36 rotationally
variant features. So, the resulting feature vector is of 32 + 32 + 32 = 96 dimension
when we apply DFT normalization to achieve rotationally invariant features.
Gabor Features + Chromatic Features: Different frequencies and orientations are
used in Gabor filters for extracting features from an image. The tuning of Gabor
filters with these parameters: number of orientations (no) = 6, number of frequen-
cies (nF) = 4, frequency ratio (Fr) = half-octave, standard deviation of Gaussian
envelope (longitudinal) ( ) = 0.5, standard deviation of Gaussian envelop
(transversal) ( ) = 0.5, max frequency (Fm) = 0.327. This tuning is same in overall
experiment.
Opponent Color LBP: The processing of opponent color LBP feature extraction
method is same as LBP method. In opponent color LBP method, we calculate features
from inter-channel and intra-channel, but in LBP method, we calculate features from
grayscale image. Separate channels give 108 (36 + 36 + 36) features, and paired
channels give 108 features. The dimension of resulting feature vector is 216.
Color Ranklets: Ranklet transform is an image processing technique which
considers the relative rank of neighboring pixels instead of their values. Invariance
against rotation is achieved by computing the discrete Fourier transform (DFT) of a
set of ranklet features computed from circular neighborhoods at different rotation
angles.
6 M. Ritesh and S. Ashwani](https://image.slidesharecdn.com/26279-250427032823-aa6dc0e6/85/Sensors-and-Image-Processing-Proceedings-of-CSI-2015-1st-Edition-Shabana-Urooj-23-320.jpg)
![2.3 Classifier
Nearest Neighbor and Nearest Mean Classifier: An input pattern is classified in the
class of the nearest training pattern (NN) or that of the nearest centroid (NMC) [37,
38]. Absence of parameter tuning makes these classifiers advantageous and easy to
implement. In NN, sensitivity to outliers makes it poor performer. In nearest mean
classifier, misclassifications arise because probably centroids are not representative.
Linear Classifier: Linear classifiers classify features by making a classification
decision which is based on linear combination of the feature values. Linear classifier is
originally developed for binary classification. It requires a predefined linear function
(hyperplane) that best separates the required classes in the feature space. If the two
classes are linearly separable, then perfect separation between classes exists.
Support Vector Classifier: support vector machine is highly effective classifier
and currently has a great importance in pattern recognition and artificial intelli-
gence. The tuning of a SVC is very important and needs very careful analysis. In
our experiment, we are using RBF kernel function in SVM classifier; then, we have
to tune two parameters: C and gamma (the radius of RBF).
C is used during the training phase and says how much outliers are taken into
account in calculating support vector. C is a trade-off between training error and the
flatness of the solution. If the value of C is larger, then the final training error will be
less. But if we increase value of C too much, then the risk of losing the general-
ization properties of the classifier is high. Large C increases the time needed for
training, and small C makes classifier flat. We have to find a C that keeps the
training error small and generalizes well. In SVC processing, we choose a kernel
function which mapped patterns into a high-dimensional space [39]. According to
Hsu et al. [40], if the number of features is not large, then RBF kernel is a rea-
sonable choice.
3 Experiments and Simulation
3.1 Database Preparation
In our experiment, we use database of skin cancer images (melanoma and
non-melanoma) have been used from University of Waterloo. The University has
collected these images from Internet source [41, 42]. Collection of 150 images is
used in this experiment. Seventy-five images are of melanoma and 75 of
non-melanoma type [43, 44].
Size of image is always an important aspect of image processing experiments. In
this experiment, input image is resized to 128*
128, which is quite suitable reso-
lution for obtaining better results.
Resized images are segmentated using k-means clustering, which are trained and
tested on different classifiers.
A Comparative Study of Various Color Texture Features for Skin … 7](https://image.slidesharecdn.com/26279-250427032823-aa6dc0e6/85/Sensors-and-Image-Processing-Proceedings-of-CSI-2015-1st-Edition-Shabana-Urooj-24-320.jpg)
Sensors and Image Processing Proceedings of CSI 2015 1st Edition Shabana Urooj
- 1. Sensors and Image Processing Proceedings of CSI 2015 1st Edition Shabana Urooj download https://textbookfull.com/product/sensors-and-image-processing- proceedings-of-csi-2015-1st-edition-shabana-urooj/ Download more ebook from https://textbookfull.com
- 2. We believe these products will be a great fit for you. Click the link to download now, or visit textbookfull.com to discover even more! Speech and Language Processing for Human Machine Communications Proceedings of CSI 2015 1st Edition S. S. Agrawal https://textbookfull.com/product/speech-and-language-processing- for-human-machine-communications-proceedings-of-csi-2015-1st- edition-s-s-agrawal/ Silicon Photonics amp High Performance Computing Proceedings of CSI 2015 1st Edition Anurag Mishra https://textbookfull.com/product/silicon-photonics-amp-high- performance-computing-proceedings-of-csi-2015-1st-edition-anurag- mishra/ Computational Methods and Clinical Applications for Spine Imaging Third International Workshop and Challenge CSI 2015 Held in Conjunction with MICCAI 2015 Munich Germany October 5 2015 Proceedings 1st Edition Tomaž Vrtovec https://textbookfull.com/product/computational-methods-and- clinical-applications-for-spine-imaging-third-international- workshop-and-challenge-csi-2015-held-in-conjunction-with- miccai-2015-munich-germany-october-5-2015-proceedings-1st/ Applied Computer Vision and Image Processing Proceedings of ICCET 2020 Volume 1 Brijesh Iyer https://textbookfull.com/product/applied-computer-vision-and- image-processing-proceedings-of-iccet-2020-volume-1-brijesh-iyer/
- 3. Proceedings of the 2015 International Conference on Communications Signal Processing and Systems 1st Edition Qilian Liang https://textbookfull.com/product/proceedings-of- the-2015-international-conference-on-communications-signal- processing-and-systems-1st-edition-qilian-liang/ Frontiers in Materials Processing Applications Research and Technology Select Proceedings of FiMPART 2015 1st Edition M. Muruganant https://textbookfull.com/product/frontiers-in-materials- processing-applications-research-and-technology-select- proceedings-of-fimpart-2015-1st-edition-m-muruganant/ Image Operators: Image Processing in Python 1st Edition Jason M. Kinser https://textbookfull.com/product/image-operators-image- processing-in-python-1st-edition-jason-m-kinser/ Advances in Image and Graphics Technologies 10th Chinese Conference IGTA 2015 Beijing China June 19 20 2015 Proceedings 1st Edition Tieniu Tan https://textbookfull.com/product/advances-in-image-and-graphics- technologies-10th-chinese-conference-igta-2015-beijing-china- june-19-20-2015-proceedings-1st-edition-tieniu-tan/ Image operators image processing in Python First Edition Kinser https://textbookfull.com/product/image-operators-image- processing-in-python-first-edition-kinser/
- 4. Advances in Intelligent Systems and Computing 651 Sensors and Image Processing Shabana Urooj JitendraVirmani Editors Proceedings of CSI 2015
- 5. Advances in Intelligent Systems and Computing Volume 651 Series editor Janusz Kacprzyk, Polish Academy of Sciences, Warsaw, Poland e-mail: kacprzyk@ibspan.waw.pl
- 6. About this Series 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. The publications within “Advances in Intelligent Systems and Computing” are primarily textbooks and 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. Advisory Board Chairman Nikhil R. Pal, Indian Statistical Institute, Kolkata, India e-mail: nikhil@isical.ac.in Members Rafael Bello Perez, Universidad Central “Marta Abreu” de Las Villas, Santa Clara, Cuba e-mail: rbellop@uclv.edu.cu Emilio S. Corchado, University of Salamanca, Salamanca, Spain e-mail: escorchado@usal.es Hani Hagras, University of Essex, Colchester, UK e-mail: hani@essex.ac.uk László T. Kóczy, Széchenyi István University, Győr, Hungary e-mail: koczy@sze.hu Vladik Kreinovich, University of Texas at El Paso, El Paso, USA e-mail: vladik@utep.edu Chin-Teng Lin, National Chiao Tung University, Hsinchu, Taiwan e-mail: ctlin@mail.nctu.edu.tw Jie Lu, University of Technology, Sydney, Australia e-mail: Jie.Lu@uts.edu.au Patricia Melin, Tijuana Institute of Technology, Tijuana, Mexico e-mail: epmelin@hafsamx.org Nadia Nedjah, State University of Rio de Janeiro, Rio de Janeiro, Brazil e-mail: nadia@eng.uerj.br Ngoc Thanh Nguyen, Wroclaw University of Technology, Wroclaw, Poland e-mail: Ngoc-Thanh.Nguyen@pwr.edu.pl Jun Wang, The Chinese University of Hong Kong, Shatin, Hong Kong e-mail: jwang@mae.cuhk.edu.hk More information about this series at http://www.springer.com/series/11156
- 7. Shabana Urooj • Jitendra Virmani Editors Sensors and Image Processing Proceedings of CSI 2015 123
- 8. Editors Shabana Urooj Gautam Buddha University Greater Noida, Uttar Pradesh India Jitendra Virmani Department of Electrical and Instrumentation Engineering Thapar University Patiala, Punjab India ISSN 2194-5357 ISSN 2194-5365 (electronic) Advances in Intelligent Systems and Computing ISBN 978-981-10-6613-9 ISBN 978-981-10-6614-6 (eBook) https://doi.org/10.1007/978-981-10-6614-6 Library of Congress Control Number: 2017952512 © Springer Nature Singapore Pte Ltd. 2018 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, express 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. Printed on acid-free paper This Springer imprint is published by Springer Nature The registered company is Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore
- 9. Preface The last decade has witnessed remarkable changes in IT industry, virtually in all domains. The 50th Annual Convention, CSI-2015, on the theme “Digital Life” was organized as a part of CSI-2015, by CSI at Delhi, the national capital of the country, during December 02–05, 2015. Its concept was formed with an objective to keep ICT community abreast of emerging paradigms in the areas of computing tech- nologies and more importantly looking at its impact on the society. Information and Communication Technology (ICT) comprises of three main components: infrastructure, services, and product. These components include the Internet, infrastructure-based/infrastructure-less wireless networks, mobile termi- nals, and other communication mediums. ICT is gaining popularity due to rapid growth in communication capabilities for real-time-based applications. “Nature Inspired Computing” is aimed at highlighting practical aspects of computational intelligence including robotics support for artificial immune systems. CSI-2015 attracted over 1500 papers from researchers and practitioners from academia, industry, and government agencies, from all over the world, thereby making the job of the Programme Committee extremely difficult. After a series of tough review exercises by a team of over 700 experts, 565 papers were accepted for presentation in CSI-2015 during the 3 days of the convention under ten parallel tracks. The Programme Committee, in consultation with Springer, the world’s largest publisher of scientific documents, decided to publish the proceedings of the presented papers, after the convention, in ten topical volumes, under ASIC series of the Springer, as detailed hereunder: 1. Volume # 1: ICT Based Innovations 2. Volume # 2: Next Generation Networks 3. Volume # 3: Nature Inspired Computing 4. Volume # 4: Speech and Language Processing for Human-Machine Communications 5. Volume # 5: Sensors and Image Processing 6. Volume # 6: Big Data Analytics v
- 10. 7. Volume # 7: Systems and Architecture 8. Volume # 8: Cyber Security 9. Volume # 9: Software Engineering 10. Volume # 10: Silicon Photonics and High Performance Computing We are pleased to present before you the proceedings of Volume # 5 on “Sensors and Image Processing.” The title “Sensors and Image Processing” highlights the different applications in the field of virtual reality. It also delves into the matter as to how robotics can be applied to strengthen modeling. The title also showcases the various augments of latest mobile technologies, solid modeling, etc. Sensors are used in everyday objects such as touch-sensitive elevator buttons and lamps which dim or brighten by touching the base, besides innumerable applications of which most people are never aware of. With advances in micro- machinery and easy-to-use microcontroller platforms, the uses of sensors have expanded beyond the more traditional fields of temperature, pressure, or flow measurement, for example MARG sensors. Image processing is processing of images using mathematical operations by using any form of signal processing for which the input is an image, such as a photograph or video frame. The output of image processing may be either an image or a set of characteristics or parameters related to the image. Most image-processing techniques involve treating the image as a two-dimensional signal and applying standard signal-processing techniques to it. The title “Sensors and Image Processing” also amalgamates and showcases the applications of above technologies in different research and real-time domains. The volume includes scientific, original, and high-quality papers presenting novel research, ideas, and explorations of new vistas in speech and language processing such as speech recognition, text recognition, embedded platform for information retrieval, segmentation, filtering and classification of data, emotion recognition. The aim of this volume is to provide a stimulating forum for sharing knowledge and results in the model, methodology, and implementations of speech and language processing tools. Its authors are researchers and experts of these domains. This volume is designed to bring together researchers and practitioners from academia and industry to focus on extending the understanding and establishing new col- laborations in these areas. It is the outcome of the hard work of the editorial team, who have relentlessly worked with the authors and steered up the same to compile this volume. It will be a useful source of reference for the future researchers in this domain. Under the CSI-2015 umbrella, we received over 200 papers for this vol- ume, out of which 29 papers are being published, after rigorous review processes, carried out in multiple cycles. On behalf of organizing team, it is a matter of great pleasure that CSI-2015 has received an overwhelming response from various professionals across the country. The organizers of CSI-2015 are thankful to the members of Advisory Committee, Programme Committee, and Organizing Committee for their all-round guidance, encouragement, and continuous support. We express our sincere gratitude to the learned Keynote Speakers for support and help extended to make this event a grand success. Our sincere thanks are also due to our Review Committee Members and the vi Preface
- 11. Editorial Board for their untiring efforts in reviewing the manuscripts, giving suggestions and valuable inputs for shaping this volume. We hope that all the participated delegates will be benefitted academically and wish them for their future endeavors. We also take the opportunity to thank the entire team from Springer, who have worked tirelessly and made the publication of the volume a reality. Last but not least, we thank the team from Bharati Vidyapeeth’s Institute of Computer Applications and Management (BVICAM), New Delhi, for their untiring support, without which the compilation of this huge volume would not have been possible. Greater Noida, India Shabana Urooj Patiala, India Jitendra Virmani March 2017 Preface vii
- 12. The Organization of CSI-2015 Chief Patron Padmashree Dr. R. Chidambaram, Principal Scientific Advisor, Government of India Patrons Prof. S.V. Raghavan, Department of Computer Science, IIT Madras, Chennai Prof. Ashutosh Sharma, Secretary, Department of Science and Technology, Ministry of Science and Technology, Government of India Chair, Programme Committee Prof. K.K. Aggarwal, Founder Vice Chancellor, GGSIP University, New Delhi Secretary, Programme Committee Prof. M.N. Hoda, Director, Bharati Vidyapeeth’s Institute of Computer Applications and Management (BVICAM), New Delhi Advisory Committee • Padma Bhushan Dr. F.C. Kohli, Co-Founder, TCS • Mr. Ravindra Nath, CMD, National Small Industries Corporation, New Delhi • Dr. Omkar Rai, Director General, Software Technological Parks of India (STPI), New Delhi • Adv. Pavan Duggal, Noted Cyber Law Advocate, Supreme Courts of India • Prof. Bipin Mehta, President, CSI • Prof. Anirban Basu, Vice President—cum- President Elect, CSI • Shri Sanjay Mohapatra, Secretary, CSI • Prof. Yogesh Singh, Vice Chancellor, Delhi Technological University, Delhi • Prof. S.K. Gupta, Department of Computer Science and Engineering, IIT, Delhi ix
- 13. • Prof. P.B. Sharma, Founder Vice Chancellor, Delhi Technological University, Delhi • Mr. Prakash Kumar, IAS, Chief Executive Officer, Goods and Services Tax Network (GSTN) • Mr. R.S. Mani, Group Head, National Knowledge Networks (NKN), NIC, Government of India, New Delhi Editorial Board • M.U. Bokhari, AMU, Aligarh • D.K. Lobiyal, JNU, New Delhi • Umang Singh, ITS, Ghaziabad • Shiv Kumar, CSI • Vishal Jain, BVICAM, New Delhi • Shalini Singh Jaspal, BVICAM, New Delhi • S.S. Agrawal, KIIT, Gurgaon • Amita Dev, BPIBS, New Delhi • Aasim Zafar, AMU, Aligarh • Ritika Wason, BVICAM, New Delhi • Anupam Baliyan, BVICAM, New Delhi • S.M.K. Quadri, JMI, New Delhi x The Organization of CSI-2015
- 14. Contents A Comparative Study of Various Color Texture Features for Skin Cancer Detection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Maurya Ritesh and Srivastava Ashwani Pattern Classification and Retrieval of Content-Based Images—a Step Towards Amelioration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Hemjot and Amitabh Sharma Novel Threshold Formulation for Energy Detection Method to Efficient Spectrum Sensing in Cognitive Radio. . . . . . . . . . . . . . . . . . . . . 25 Rohini S. Kale, Vijay M. Wadhai and Jagdish B. Helonde A Novel Architecture to Crawl Images Using OAI-PMH. . . . . . . . . . . . . 37 Shruti Sharma, Parul Gupta and C.K. Nagpal A Novel Web Page Change Detection Technique for Migrating Crawlers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 Ashlesha Gupta, Ashutosh Dixit and A.K. Sharma Developing Prototype for Prosopagnosia Using PCA . . . . . . . . . . . . . . . . 59 Gunjan Jhawar, Prajacta Nagraj and N. Ramesh Babu Implementation of Smart Video Surveillance System Using Motion Detection Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 Gaurav Verma, Swati Gautam, Rohan Agarwal, Sarthak Saxena and Deval Verma Intelligent Algorithm for Automatic Multistoried Parking System Using Image Processing with Vehicle Tracking and Monitoring from Different Locations in the Building . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 Sagar Juneja, Saurav Kochar and Sachin Dhiman Computer-Aided Detection of Breast Cancer Using Pseudo Zernike Moment as Texture Descriptors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 Shabana Urooj, Satya P. Singh and A.Q. Ansari xi
- 15. Analysis of Leukoderma Images Using Neuro-Fuzzy Hybrid Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 Sudhakar Singh, Shabana Urooj and Satya P. Singh Texture and Clustering-based Skin Disease Classification . . . . . . . . . . . . 103 Pradeep Mullangi, Y. Srinivasa Rao and Pushpa Kotipalli Melanoma Skin Cancer Detection Using Image Processing . . . . . . . . . . . 111 Nishtha Garg, Vishakha Sharma and Prabhjot Kaur Performance Assessment for Clustering Techniques for Image Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 Jyoti Arora, Kiran Khatter and Meena Tushir Automatic Seed Point Selection in B-Mode Breast Ultrasound Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 Madan Lal and Lakhwinder Kaur EEA-LEACH—Trustworthy-Enhanced Algorithm for Energy Optimization in Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . 139 Umang, M.N. Hoda and Geetanjali Singh Secure Cluster-Based Data Integrity Routing for Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 Shashikala and C. Kavitha Objects Detection and Tracking Strategies . . . . . . . . . . . . . . . . . . . . . . . . 169 Sureshbabu Matla and Ravi Subban Road Accident Informer System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181 S. Vanitha Hybrid Intelligent Algorithm for Energy-Efficient Routing in WSN . . . . . . . . 189 Raminder Singh Uppal A Method for Secret Image Transmission to Preserve Privacy . . . . . . . . 201 Rasika Thakare, B Sumit and Umesh Kulkarni Text Retrieval from Scanned Forms Using Optical Character Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207 Vaishali Aggarwal, Sourabh Jajoria and Apoorvi Sood Combined Approach of Fuzzy Image Blocking Implementation for Medical Image Fusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217 Madhuri Patel and Drashti Baldev A Systematic Review on Image Enhancement Techniques. . . . . . . . . . . . 227 Nishant Bhardwaj, Gagandeep Kaur and Pradeep Kumar Singh Virtual Field and Wave Laboratory Using Scilab and Java . . . . . . . . . . 237 Sarika Agarwal and Gunjeet Kaur xii Contents
- 16. Classification and Comparative Study of IRS LISS-III Satellite Images Using RBFN and Decision Tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245 Anand Upadhyay, Santosh kumar Singh, Shailesh Kisan Gaikwad and Ashutosh Chandra Mukherjee An Analytic Review on Image Enhancement Techniques Based on Soft Computing Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255 Gagandeep Kaur, Nishant Bhardwaj and Pradeep Kumar Singh An Auto-Threshold Control for Isolating False Shadow from High-Resolution Images. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267 K. Komathy Contents xiii
- 17. About the Editors Dr. Shabana Urooj currently works at Gautam Buddha University, Greater Noida, India. She received her bachelor’s in Electrical Engineering and M.Tech. in Instrumentation and Control from Aligarh Muslim University, Aligarh, India. She has completed her doctorate at Jamia Millia Islamia (A Central University), New Delhi, India. She has 3 years of industrial and 15 years of teaching experience. She has authored/co-authored more than 90 research papers. She has served as an editor, board member, and reviewer for a number of leading journals. She is an active volunteer of IEEE and actively involved in various professional development ventures and bodies, such as the CSI and the ISTE. Jitendra Virmani received his B.Tech. (Hons) in Instrumentation Engineering from Sant Longowal Institute of Engineering and Technology, Punjab, in 1999; his M.Tech. in Electrical Engineering with specialization in measurement and instru- mentation from the Indian Institute of Technology Roorkee, India, in 2006; and his Ph.D. on Analysis and Classification of B-Mode Liver Ultrasound Images from the same institute in 2014. He held various academic posts before joining the Department of Electrical and Instrumentation Engineering, Thapar University, Patiala, Punjab, India, where he has been an Assistant Professor since July 2015. He is a life member of the Institute of Engineers (IEI), India. His research interests include application of machine learning and soft computing techniques for the analysis of medical images. xv
- 18. A Comparative Study of Various Color Texture Features for Skin Cancer Detection Maurya Ritesh and Srivastava Ashwani Abstract Detection of skin cancer gives the best chance of being diagnosed early. Biopsy method for skin cancer detection is much painful. Human interpretation contains difficulty and subjectivity; therefore, automated analysis of skin cancer-affected images has become important. This paper proposes an automatic medical image classification method to classify two major type skin cancers: melanoma and non-melanoma. In this paper, we have used the color and texture features in combination which gives better results than using color or gray-level information alone. We have used k-means clustering algorithm to segment the lesion. The features are extracted by seven different color texture feature extractors from the segmented images. Classification accuracy of our proposed system is evaluated on four different types of classifiers, and their values are compared with one another. The results of the proposed system are computed on fivefolds of cross-validation in order to perform better analysis of our proposed method. Keywords Gray-level co-occurrence matrices Support vector machine Local binary patterns Texture features Color percentiles K-means clustering Co-occurrence matrix Color features Integrative co-occurrence matrix Gabor features Linear classifier NN classifiers NMC classifiers Cross-validation M. Ritesh () Department of Computer Science and Engineering, Shri Ramswaroop Memorial University, Lucknow, U.P, India e-mail: maurya123ritesh47@gmail.com S. Ashwani Computer Engineering and Application Department, GLA University, Mathura, India e-mail: ashwani.sri89@gmail.com © Springer Nature Singapore Pte Ltd. 2018 S. Urooj and J. Virmani (eds.), Sensors and Image Processing, Advances in Intelligent Systems and Computing 651, https://doi.org/10.1007/978-981-10-6614-6_1 1
- 19. 1 Introduction Skin cancer incidence is increasing at 3.1% per year [1]. Skin cancer spread over the body with the help of lymphatic and blood vessels. Thus, early detection of skin cancer is very important for proper diagnosis of the disease. Melanoma and non-melanoma are two major categories of skin cancers. Malignant melanoma is of several subtypes. Basal cell carcinoma and squamous cell carcinomas are two main types of non-melanoma skin cancers. Each type of skin cancer is different from the other skin cancers in certain characteristics. In clinical detection of skin cancer diagnosis, dermatologists use a visual inspection. Clinical diagnostic performance is very poor in comparison with der- moscopy and automatic diagnosis. Dermoscopy is a noninvasive diagnostic tech- nique. It uses clinical dermatology and dermatopathology in combination to inspect the morphological features which is not possible in clinical detection. Dermoscopy increases the performance of diagnosis with 10–30% compared to unaided eye [2]. Differentiation of skin cancer images needs much more experience with der- moscopy technique. Less-experienced clinicians use ABCD-E rule [3] to improve the diagnostic performance [3]. Automatic image processing of skin cancer gives better results by providing the exact information about lesion, which can be useful for the clinician to detect and classify skin cancer. It is also used as a stand-alone early warning tool. Effective implementation of this automatic technique may give reduced deaths with benefits both to the patients and to the healthcare system. Working of automatic skin cancer detection has three main stages: (1) segmentation of lesion; (2) feature extraction and feature selection; and (3) lesion classification. Segmentation is an important process in image processing applications and computer vision because doctors are always interested in meaningful regions of the dermoscopy image. Segmentation divides an image into a number of separate regions. Pixels, in each region, have high similarity such as color, intensity, and texture. Many researchers use only gray level for image segmentation [31]. But, in our proposed system, we use color information of the image for lesion segmentation. In general, we covert the color image in gray-level image; therefore, color information does not used. There is a wide variety of segmentation methods used in dermoscopy images [4]. Recent advancements include thresholding [5, 6], k-means clustering [7], fuzzy c-means clustering [8, 9], density-based clustering [10], mean-shift clustering [11], gradient vector flow snakes [12–14], color quantization followed by spatial segmentation [15], statistical region merging [16], watershed transformation [17], dynamic programming [18, 19], and supervised learning [20, 21]. Clustering is an unsupervised learning technique, where one can give the number of clusters in advance to classify pixels [22]. A similarity measure is defined between pixels, and similar pixels are then grouped into a cluster. We use k-means clustering for segmentation of color images. 2 M. Ritesh and S. Ashwani
- 20. It is very hard to differentiate skin cancer visually. Identification and extraction of most effective features from cancer-affected lesion is very important. Each class of skin cancer has some different features than others. We use these different features for classification. Feature extraction extracts useful features or properties from original dataset of an image. These extracted features easily classify the classes of skin cancer. Color features are mainly statistical parameters. These are calculated from inter- and intra-channel of an image, like average value and standard deviation of the RGB [23–27] or HSV color channels [28]. Here, we use “local binary pat- terns + color percentiles,” “integrative co-occurrence matrices,” “gray-level co-occurrence matrices + color percentiles,” “Gabor features + chromatic fea- tures,” “Gabor features,” “opponent color LBP,” “color ranklets” [29]. These methods are based on texture and disjoint color analysis. Textural features are extracted from images by converting into gray level, and color features are com- puted with the help of three color component of an image. Textural and color features are concatenated into the same feature vector to improve the classification accuracy. The main aim of feature selection is to select the maximum number of features to achieve high performance in cancer classification [30]. Feature selection is important when anyone works on gray-level features. In our proposed system, there is no need of feature selection algorithms. Classification phase of the diagnostic system is the one in charge of making the inferences about the extracted information in the previous phases in order to be able to produce a diagnostic about the input image. In our experiment, we have used four well-established classifiers: support vector classifier (SVC), nearest neighborhood (NN), linear classifier, and nearest mean classifier (NMC). The rest of the paper is organized as follows: Sect. 2 briefly reviews segmen- tation, feature extraction techniques, and classification methods which are used in proposed framework. Section 3 reports extensive experimental results, and Sect. 4 concludes the paper. 2 Proposed Framework Proposed framework is a compiled abstraction of digital image classification. Figure 1 describes the steps in an image classification. In this paper, we propose this framework with k-means clustering segmentation and color texture feature extraction techniques which is a new approach to classify the skin cancer images. For validation purpose, twofold cross-validation is used. A Comparative Study of Various Color Texture Features for Skin … 3
- 21. 2.1 Segmentation Segmentation is a process to partitioning an image into disjoint regions that are homogeneous with respect to a chosen property such as luminance, color, and texture. The aim of segmentation is to change the representation of an image into something that is more meaningful and easier to analyze [30]. K-means Clustering Segmentation: K-means clustering method is one of the simplest unsupervised learning methods and this method is nondeterministic, numerical, and iterative. In this experiment, color images are used as input; hence, Fig. 1 Proposed method 4 M. Ritesh and S. Ashwani
- 22. this technique is used for segmentation. K-means clustering is partitioning method. This method groups objects in the way that within-group variance is minimized. If within-group variance is minimized, then it gives high featured segmented image. The working of this method is as follows [31, 32]: 1. Select K pixels for centers randomly. Initially, center represents group centroids. 2. For grouping the sample data, calculate histogram bin value distance between pixels and selected centroids and assign the group on the basis of nearest distance. 3. Calculate the histogram bin value for new group to find the new position of centroids. 4. If the value of centroids changes, then repeat steps 2 and 3 (Fig. 2). 2.2 Feature Extraction It is important to identify the most effective features to extract from skin cancer lesions and to find the most effective pattern-classification criteria and algorithms for differentiating those lesions. Local Binary Patterns + Color Percentiles: The LBP feature vector is created in the following way: 1. Divide the window size into cells in the form of n* n matrix (i.e., 16* 16). 2. If the center pixel value (Threshold Value) is less than the selected pixel, then write 1, otherwise write 0. For rotational invariant features, move in one direction either clockwise or anticlockwise. This step gives 8-digit binary number. Convert it into decimal number for further processing. 3. Compute the histogram of the frequency of each “number” occurring over the given cell. 4. Apply normalization on histograms. 5. Concatenate the histograms of all divided cells. The combination of LBP + CP presented by Niskanen et al. [33]. In this method, we calculate local binary patterns in an grayscale image. For better results, we reduce the calculated features into rotationally invariant features. Rotational invariance is necessary because when image is rotated, the gray values also rotated in a circular form where origin is same. Then, calculation of feature vector changes Fig. 2 Input image and segmented image A Comparative Study of Various Color Texture Features for Skin … 5
- 23. with the rotation. We apply discrete Fourier transformation function (DFT) to make features rotationally invariant. There are always 36 rotationally invariant features in a grayscale image. Now, we calculate the color percentiles of first, second, and third quartiles (three points that divide the each channel into four parts) of three channels. Total features from color percentiles are nine. When we concatenate features of these two methods, the final resulting feature vector dimension is 36. Integrative Co-occurrence Matrices: Integrative co-occurrence matrix [34–36] is a new approach for calculating the inter-channel and intra-channel features. In inter-channel feature calculation, we calculate the co-occurrence matrix features from each of three channels, and in intra-channel, we calculate the features from combination of color channels ((r,g), (r,b), (g,b)). In our experiment, we use five co-occurrence features: energy, contrast, correlation, homogeneity, and inertia. Final resulting feature vector dimension is 30. Co-occurrence Matrices + Color Percentiles: The combination of co-occurrence matrices and color percentiles is presented by Niskanen et al. [33]. This method is first used for applications in wood inspection. Out of 14 features, we computed five important co-occurrence features for classification. Color percentile is calculated as discussed in LBPnCP method. So, the dimension offinal resulting feature vector is 14. Gabor Features: Gabor features are used to extract scale and orientation infor- mation from different channels. We calculate Gabor features for inter-channel of an image with above-mentioned configuration. One channel gives 36 rotationally variant features. So, the resulting feature vector is of 32 + 32 + 32 = 96 dimension when we apply DFT normalization to achieve rotationally invariant features. Gabor Features + Chromatic Features: Different frequencies and orientations are used in Gabor filters for extracting features from an image. The tuning of Gabor filters with these parameters: number of orientations (no) = 6, number of frequen- cies (nF) = 4, frequency ratio (Fr) = half-octave, standard deviation of Gaussian envelope (longitudinal) ( ) = 0.5, standard deviation of Gaussian envelop (transversal) ( ) = 0.5, max frequency (Fm) = 0.327. This tuning is same in overall experiment. Opponent Color LBP: The processing of opponent color LBP feature extraction method is same as LBP method. In opponent color LBP method, we calculate features from inter-channel and intra-channel, but in LBP method, we calculate features from grayscale image. Separate channels give 108 (36 + 36 + 36) features, and paired channels give 108 features. The dimension of resulting feature vector is 216. Color Ranklets: Ranklet transform is an image processing technique which considers the relative rank of neighboring pixels instead of their values. Invariance against rotation is achieved by computing the discrete Fourier transform (DFT) of a set of ranklet features computed from circular neighborhoods at different rotation angles. 6 M. Ritesh and S. Ashwani
- 24. 2.3 Classifier Nearest Neighbor and Nearest Mean Classifier: An input pattern is classified in the class of the nearest training pattern (NN) or that of the nearest centroid (NMC) [37, 38]. Absence of parameter tuning makes these classifiers advantageous and easy to implement. In NN, sensitivity to outliers makes it poor performer. In nearest mean classifier, misclassifications arise because probably centroids are not representative. Linear Classifier: Linear classifiers classify features by making a classification decision which is based on linear combination of the feature values. Linear classifier is originally developed for binary classification. It requires a predefined linear function (hyperplane) that best separates the required classes in the feature space. If the two classes are linearly separable, then perfect separation between classes exists. Support Vector Classifier: support vector machine is highly effective classifier and currently has a great importance in pattern recognition and artificial intelli- gence. The tuning of a SVC is very important and needs very careful analysis. In our experiment, we are using RBF kernel function in SVM classifier; then, we have to tune two parameters: C and gamma (the radius of RBF). C is used during the training phase and says how much outliers are taken into account in calculating support vector. C is a trade-off between training error and the flatness of the solution. If the value of C is larger, then the final training error will be less. But if we increase value of C too much, then the risk of losing the general- ization properties of the classifier is high. Large C increases the time needed for training, and small C makes classifier flat. We have to find a C that keeps the training error small and generalizes well. In SVC processing, we choose a kernel function which mapped patterns into a high-dimensional space [39]. According to Hsu et al. [40], if the number of features is not large, then RBF kernel is a rea- sonable choice. 3 Experiments and Simulation 3.1 Database Preparation In our experiment, we use database of skin cancer images (melanoma and non-melanoma) have been used from University of Waterloo. The University has collected these images from Internet source [41, 42]. Collection of 150 images is used in this experiment. Seventy-five images are of melanoma and 75 of non-melanoma type [43, 44]. Size of image is always an important aspect of image processing experiments. In this experiment, input image is resized to 128* 128, which is quite suitable reso- lution for obtaining better results. Resized images are segmentated using k-means clustering, which are trained and tested on different classifiers. A Comparative Study of Various Color Texture Features for Skin … 7
- 25. 3.2 Result Analysis The size of feature vector extracted using these feature extraction algorithms is shown below in Table 1. It is clear from the table increasing the dimensionality of feature vector increases the computation time (Fig. 3). Tables 2, 3, 4, 5, 6, 7, and 8 display classification accuracy of different classifiers for the given feature set using twofold cross-validation approach by varying the training and testing ratio. Table 1 Elapsed time of methods in increasing order Method Number of features Elapsed time (s) Reference LBPnCP 45 54.533748 (M. Niskanen et al. 2001) GLCM + CP 14 90.964714 (M. Niskanen et al. 2001) Gabor + CF 36 110.209004 (A. Drimbarean et al. 2001) Gabor 96 208.859732 (G. Paschos, 2001) Integrated CM 30 393.141803 (V. Arvis et al. 2004) Opponent color LBP 216 13781.195905 (T. Maenpaa et al. 2004) Color ranklets 216 28463.284710 (F. Bianconi et al. 2009) Fig. 3 Images from dataset first row: melanoma, second row: non-melanoma Table 2 Accuracy of LBPnCP feature extraction technique on different classifiers with given training and testing ratio Training/testing NN NMC Linear SVC (60,15) 65.56 50.00 63.33 62.00 (55,20) 57.14 61.43 51.43 57.14 (50,25) 68.00 58.00 62.00 66.00 (45,30) 80.00 66.67 80.00 83.33 (40,35) 80.00 70.00 93.36 89.14 8 M. Ritesh and S. Ashwani
- 26. Table 3 Accuracy of Gabor + CF feature extraction technique on different classifiers with given training and testing ratio Training/testing NN NMC Linear SVC (60,15) 62.00 66.67 63.33 68.89 (55,20) 70.00 77.14 70.00 68.57 (50,25) 76.00 80.00 70.00 82.00 (45,30) 86.67 46.67 86.67 80.00 (40,35) 90.00 80.00 80.00 90.00 Table 4 Accuracy of integrated CM feature extraction technique on different classifiers with given training and testing ratio Training/testing NN NMC Linear SVC (60,15) 70.00 66.67 53.33 66.67 (55,20) 64.29 62.86 67.14 70.00 (50,25) 84.00 80.00 74.00 72.00 (45,30) 86.67 83.33 83.33 86.67 (40,35) 90.00 70.00 80.00 87.76 Table 5 Accuracy of GLCM + CP feature extraction technique on different classifiers with given training and testing ratio Training/testing NN NMC Linear SVC (60,15) 60.00 57.78 58.89 64.44 (55,20) 61.43 47.14 68.57 64.29 (50,25) 74.00 84.00 82.00 82.00 (45,30) 80.00 76.67 83.33 83.33 (40,35) 91.24 60.00 88.65 93.94 Table 6 Accuracy of Gabor feature extraction technique on different classifiers with given training and testing ratio Training/testing NN NMC Linear SVC (60,15) 71.11 74.44 74.44 76.67 (55,20) 64.29 71.43 54.29 67.14 (50,25) 78.00 74.00 78.00 78.00 (45,30) 90.00 83.33 76.67 90.00 (40,35) 90.00 80.00 60.00 96.57 Table 7 Accuracy of opponent color LBP feature extraction technique on different classifiers with given training and testing ratio Training/testing NN NMC Linear SVC (60,15) 55.56 65.56 60.00 63.33 (55,20) 67.14 68.57 64.29 68.57 (50,25) 76.00 66.00 68.00 80.00 (45,30) 86.67 76.67 90.00 86.67 (40,35) 92.29 50.00 80.00 95.11 A Comparative Study of Various Color Texture Features for Skin … 9
- 27. Average accuracies obtained from each classifier for different features are plotted graphically in Fig. 4. It is clear from Fig. 4 that Gabor feature performs better than the other features. LBPnCP, integrated GLCM, CLCM + CP are the better choice, while the color ranklets and opponent color LBP perform poorly in time and space domains. LBPnCP and Gabor features are best suited for optimum results considering the running time and accuracy of prediction. Pairwise comparison between the algorithms is done for better visual compar- ison (Fig. 5). The directions of the arrow denote the method which has more prediction accuracy than the other. Table 8 Accuracy of color ranklets feature extraction technique on different classifiers with given training and testing ratio Training/testing NN NMC Linear SVC (60,15) 68.89 57.78 60.00 63.33 (55,20) 72.86 55.71 61.43 70.00 (50,25) 74.00 56.00 72.00 74.00 (45,30) 76.67 60.00 70.00 76.67 (40,35) 80.00 70.00 78.00 80.00 Fig. 4 Average accuracy of different feature extraction algorithms 10 M. Ritesh and S. Ashwani
- 28. 4 Conclusion Experimental results show that color and texture descriptors for skin cancer clas- sification provide good classification accuracy. We have also evaluated the per- formance of four different classifiers on these extracted features. SVC(support vector classifier) outperforms among all others with the same features set and then comes linear classifier, whose classification accuracy is nearly same as the SVC. 1-NN performs poorly on the given feature sets. Gabor features proved to be the best features that can be used for this particular application. Other more suitable methods are LBPnCP and integrated GLCM which are also giving the promising results. Opponent and color ranklets are the methods which are not advisable to be used for the particular application because of their large dimensions and less prediction accuracy as well. Fig. 5 Pairwise comparison of the feature extraction algorithm efficiency A Comparative Study of Various Color Texture Features for Skin … 11
- 29. References 1. Linos, E., Swetter, S.M., Cockburn, M.G., Colditz, G.A., Clarke, C.A.: Increasing burden of melanoma in the United States. J. Invest. Dermatol. 129(7), 1666–1674 (2009) 2. Mayer, J.: Systematic review of the diagnostic accuracy of dermatoscopy in detecting malignant melanoma. Med. J. Aust. 167(4), 206–210 (1997) 3. Robinson, J.K., Turrisi, R.: Skills training to learn discrimination of ABCDE criteria by those at risk of developing melanoma. Arch. Dermatol. 142(4), 447–452 (2006) 4. Celebi, M.E., Iyatomi, H., Schaefer, G., Stoecker, W.V.: Lesion border detection in dermoscopy images. Comput. Med. Imaging Graph. 33(2), 148–153 (2009) 5. Iyatomi, H., Oka, H., Celebi, M.E., Hashimoto, M., Hagiwara, M., Tanaka, M., Ogawa, K.: An improved internet-based melanoma screening system with dermatologist-like tumor area extraction algorithm. Comput. Med. Imaging. Graph. 32(7), 566–579 (2008) 6. Garnavi, R., Aldeen, M., Celebi, M.E., Varigos, G., Finch, S.: Border detection in dermoscopy images using hybrid thresholding on optimized color channels. Comput. Med. Imaging Graph. 35(2), 105–115 (2011) 7. H. Zhou, M. Chen, L. Zou, R. Gass, L. Ferris, L. Drogowski, and J. M. Rehg, “Spatially Constrained Segmentation of Dermoscopy Images,” Proce. 5th IEEE International Symp. Biomed. Imaging, pp. 800–803, (2008) 8. Schmid, P.: Segmentation of digitized dermatoscopic images by two-dimensional color clustering. IEEE Trans. Med. Imaging 18(2), 164–171 (1999) 9. Zhou, H., Schaefer, G., Sadka, A., Celebi, M.E.: Anisotropic mean shift based fuzzy C-means segmentation of dermoscopy images. IEEE J. Sel. Top. Sign. Proces. 3(1), 26–34 (2009) 10. Mete, M., Kockara, S., Aydin, K.: Fast density-based lesion detection in dermoscopy images. Comput. Med. Imaging Graph. 35(2), 128–136 (2011) 11. R. Melli, C. Grana, and R. Cucchiara, “Comparison of color clustering algorithms for segmentation of dermatological images,” Proc. SPIE Med. Imaging Conf. pp. 1211–1219 (2006) 12. Erkol, B., Moss, R.H., Stanley, R.J., Stoecker, W.V., Hvatum, E.: Automatic lesion boundary detection in dermoscopy images using gradient vector flow snakes. Skin Res. Technol. 11(1), 17–26 (2005) 13. Zhou, H., Schaefer, G., Celebi, M.E., Lin, F., Liu, T.: Gradient vector flow with nean shift for skin lesion segmentation. Comput. Med. Imaging Graph. 35(2), 121–127 (2011) 14. Abbas, Q., Celebi, M.E., Garcia, I.F.: A novel perceptually-oriented approach for skin tumor segmentation. Int. J. Innovative Comput. Inf. Control 8(3), 1837–1848 (2012) 15. Celebi, M.E., Aslandogan, Y.A., Stoecker, W.V., Iyatomi, H., Oka, H., Chen, X.: Unsupervised border detection in dermoscopy images. Skin Res. Technol. 13(4), 454–462 (2007) 16. Celebi, M.E., Kingravi, H.A., Iyatomi, H., Aslandogan, A., Stoecker, W.V., Moss, R.H., Malters, J.M., Grichnik, J.M., Marghoob, A.A., Rabinovitz, H.S., Menzies, S.W.: Border detection in dermoscopy images using statistical region merging. Skin Res. Technol. 14(3), 347–353 (2008) 17. Wang, H., Moss, R.H., Chen, X., Stanley, R.J., Stoecker, W.V., Celebi, M.E., Malters, J.M., Grichnik, J.M., Marghoob, A.A., Rabinovitz, H.S., Menzies, S.W., Szalapski, T.M.: Modified watershed technique and Post-Processing for segmentation of skin lesions in dermoscopy images. Comput. Med. Imaging Graph. 35(2), 116–120 (2011) 18. Abbas, Q., Celebi, M.E., Garcia, I.F., Rashid, M.: Lesion border detection in dermoscopy images using dynamic programming. Skin Res. and Technol. 17(1), 91–100 (2011) 19. Abbas, Q., Celebi, M.E., Garcia, I.F.: Skin tumor area extraction using an improved dynamic programming approach. Skin Res. Technol. 18(2), 133–142 (2012) 12 M. Ritesh and S. Ashwani
- 30. 20. Schaefer, G., Rajab, M.I., Celebi, M.E., Iyatomi, H.: Colour and contrast enhancement for improved skin lesion segmentation. Comput. Med. Imaging Graph. 35(2), 99–104 (2011) 21. Wighton, P., Lee, T.K., Lui, H., McLean, D.I., Atkins, M.S.: Generalizing common tasks in automated skin lesion diagnosis. IEEE Trans. Information Technol. Biomed. 15(4), 622–629 (2011) 22. V. K. Dehariya, S. K. Shrivastava, R. C. Jain, “Clustering of image data set using K-Means and fuzzy K-Means Algorithms”, International conference on CICN, pp. 386–391, (2010) 23. J. F. Aitken, J. Pfitzner, D. Battistutta, O.’Rourke PK, A. C. Green, and N.G. Martin, “Reliability of computer image analysis of pigmented skin lesions of Australian adolescents,” J. Cancer, 78(2), 252–257 (1996) 24. Green, A., Martin, N., Pfitzner, J., O’Rourke, M., Knight, N.: Computer image analysis in the diagnosis of melanoma. J. Am. Acad. Dermatol. 31(6), 958–964 (1994) 25. H. C. Lee, Skin cancer diagnosis using hierarchical neural networks and fuzzy logic, department of computer science, University of Missouri, Rolla (1994) 26. Seidenari, S., Burroni, M., Dell’Eva, G., Pepe, P., Belletti, B.: Computerized evaluation of pigmented skin lesion images recorded by a videomicroscope: comparison between polarizing mode observation and oil/slide mode observation. Skin Res. Technol. 1(4), 187–191 (1995) 27. K. Ramlakhan and Y. Shang,: “A mobile automated skin lesion classification system,” Proceedings of the 23rd IEEE International Conference on Tools with Artificial Intelligence (ICTAI ’11), pp. 138–141, Nov (2011) 28. Cascinelli, N., Ferrario, M., Bufalino, R., et al.: Results obtained by using a computerized image analysis system designed as an aid to diagnosis of cutaneous melanoma. Melanoma Res. 2(3), 163–170 (1992) 29. F. Bianconia, E. Gonzalezb, A. Fernandezb, S.A. Saet-taa: “Automatic classification of granite tiles through colour and texture features”, Preprint submitted to Expert Systems with Applications, Feb 20, (2012) 30. A. Masood and A. Ali Al-Jumaily, “Computer Aided Diagnostic Support System for Skin Cancer: A Review of Techniques and Algorithms” Hindawi Publishing Corporation International Journal of Biomedical Imaging, Volume (2013) 31. V. Jumb, M. Sohani, A. Shrivas: “Color image segmentation using K-Means clustering and Otsu’s adaptive thresholding,” Int. J. Innovative Technol. Exploring Eng. (IJITEE), February (2014) 32. Ojala, T., Pietikainen, M., Maenpaa, T.: Multiresolution gray-scale and rotation invariant texture classification with Local Binary Patterns. IEEE Trans. Pattern Anal. Mach. Intell. 24, 971–987 (2002) 33. Niskanen, M., Silven, O., Kauppinen, H.,: “Color and texture based wood inspection with non-supervised clustering”, Proceedings of the 12th Scandivanian Conference on Image Analysis. Bergen, pp. 336–342, (2001) 34. Arvis, V., Debain, C., Berducat, M., Benassi, A.: Generalization of the cooccurrence matrix for colour images: application to colour texture classification. Image Anal. Stereology 23(3), 63–72 (2004) 35. Palm, C.: Colour texture classification by integrative co-occurrence matrices. Pattern Recogn. 37(5), 965–976 (2004) 36. Bianconi, F., Fernandez, A.: Evaluation of the effects of Gabor filter parameters on texture classification. Pattern Recogn. 40, 3325–3335 (2007) 37. F. Bianconi, A. Fernandez, E. Gonzalez, and J. Armesto, Robust color texture features based on ranklets and discrete Fourier transform. J. Electron. Imaging 18, 043012-1-8 (2009) 38. Masotti, M., Campanini, R.: Texture classification using invariant ranklet features. Pattern Recogn. Lett. 29, 1980–1986 (2008) 39. Scholkopf, B., Smola, A.: “Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond”, MIT Press, (2002) 40. Hsu, C., Chang, C., Lin, C.: A practical guide to support vector classification. Department of Computer Science National Taiwan University, Tech. rep. (2010) A Comparative Study of Various Color Texture Features for Skin … 13
- 31. 41. http://www.dermis.net 42. http://www.dermquest.com 43. B.Gohila vani, A.Selvaraj, M.Vijaya priya, N.Anandh: “Segmentation and Classification of Skin Lesions Based on Texture Features”, Int. J. Eng. Res. Appl. Dec. (2014) 44. Maurya R., Singh S.K., Maurya A.K., Kumar A.: “GLCM and Multi Class Support vector machine based automated skin cancer classification IEEE Conference”, Page(s):444— 447Print ISBN:978-93-80544-10-6 INSPEC Accession Number:14382783, pp. 5–7 Mar. (2014) 14 M. Ritesh and S. Ashwani
- 32. Pattern Classification and Retrieval of Content-Based Images—a Step Towards Amelioration Hemjot and Amitabh Sharma Abstract The progressive web and computerized advancements have forced endless increment in the measure of visual data accessible to users. This trend prompted the advancement of exploration area where retrieval of images is done through the content of information which became familiar as CBIR (content-based image retrieval). CBIR frameworks are to a great extent utilized as a part of medicinal picture annotation, face recognition systems, security frameworks and so on. In this paper, we will discuss about an efficient system for retrieving images faster since speed and precision are important as well as techniques to obtain better classification of images. To conquer the issue of extensive number of features extracted which obliges vast measure of memory and processing force, we need to build a blend of 3 techniques (SURF, SVM and LDA) which best portray the information with adequate precision. Hence, we are using dimensionality reduction algorithm LDA in combination with SVM for the classification purpose and SURF which is quick and robust interest point detector. Keywords Content-based image retrieval (CBIR) Pattern recognition Classification Support vector machine (SVM) Colour histogram Speeded up robust features (SURF) Linear discriminant analysis (LDA) 1 Introduction A significant expansion has been done in the digital field for more than many years. This particular development has motivated research in image databases, which were virtually dismissed simply by conventional systems due to the enormous level of Hemjot () A. Sharma Department of IT, Chandigarh Engineering College, Landran, Mohali Punjab, India e-mail: hemso91@gmail.com A. Sharma e-mail: cecm.cse.amits@gmail.com © Springer Nature Singapore Pte Ltd. 2018 S. Urooj and J. Virmani (eds.), Sensors and Image Processing, Advances in Intelligent Systems and Computing 651, https://doi.org/10.1007/978-981-10-6614-6_2 15
- 33. Exploring the Variety of Random Documents with Different Content
- 34. remain freely available for generations to come. In 2001, the Project Gutenberg Literary Archive Foundation was created to provide a secure and permanent future for Project Gutenberg™ and future generations. To learn more about the Project Gutenberg Literary Archive Foundation and how your efforts and donations can help, see Sections 3 and 4 and the Foundation information page at www.gutenberg.org. Section 3. Information about the Project Gutenberg Literary Archive Foundation The Project Gutenberg Literary Archive Foundation is a non-profit 501(c)(3) educational corporation organized under the laws of the state of Mississippi and granted tax exempt status by the Internal Revenue Service. The Foundation’s EIN or federal tax identification number is 64-6221541. Contributions to the Project Gutenberg Literary Archive Foundation are tax deductible to the full extent permitted by U.S. federal laws and your state’s laws. The Foundation’s business office is located at 809 North 1500 West, Salt Lake City, UT 84116, (801) 596-1887. Email contact links and up to date contact information can be found at the Foundation’s website and official page at www.gutenberg.org/contact Section 4. Information about Donations to the Project Gutenberg Literary Archive Foundation Project Gutenberg™ depends upon and cannot survive without widespread public support and donations to carry out its mission of increasing the number of public domain and licensed works that can be freely distributed in machine-readable form accessible by the widest array of equipment including outdated equipment. Many
- 35. small donations ($1 to $5,000) are particularly important to maintaining tax exempt status with the IRS. The Foundation is committed to complying with the laws regulating charities and charitable donations in all 50 states of the United States. Compliance requirements are not uniform and it takes a considerable effort, much paperwork and many fees to meet and keep up with these requirements. We do not solicit donations in locations where we have not received written confirmation of compliance. To SEND DONATIONS or determine the status of compliance for any particular state visit www.gutenberg.org/donate. While we cannot and do not solicit contributions from states where we have not met the solicitation requirements, we know of no prohibition against accepting unsolicited donations from donors in such states who approach us with offers to donate. International donations are gratefully accepted, but we cannot make any statements concerning tax treatment of donations received from outside the United States. U.S. laws alone swamp our small staff. Please check the Project Gutenberg web pages for current donation methods and addresses. Donations are accepted in a number of other ways including checks, online payments and credit card donations. To donate, please visit: www.gutenberg.org/donate. Section 5. General Information About Project Gutenberg™ electronic works Professor Michael S. Hart was the originator of the Project Gutenberg™ concept of a library of electronic works that could be freely shared with anyone. For forty years, he produced and distributed Project Gutenberg™ eBooks with only a loose network of volunteer support.
- 36. Project Gutenberg™ eBooks are often created from several printed editions, all of which are confirmed as not protected by copyright in the U.S. unless a copyright notice is included. Thus, we do not necessarily keep eBooks in compliance with any particular paper edition. Most people start at our website which has the main PG search facility: www.gutenberg.org. This website includes information about Project Gutenberg™, including how to make donations to the Project Gutenberg Literary Archive Foundation, how to help produce our new eBooks, and how to subscribe to our email newsletter to hear about new eBooks.
- 37. back
- 38. back
- 40. back
- 41. back
- 42. back
- 43. back
- 44. back
- 45. back
- 46. back
- 47. back
- 48. back
- 49. back
- 50. back
- 52. back
- 53. Welcome to our website – the ideal destination for book lovers and knowledge seekers. With a mission to inspire endlessly, we offer a vast collection of books, ranging from classic literary works to specialized publications, self-development books, and children's literature. Each book is a new journey of discovery, expanding knowledge and enriching the soul of the reade Our website is not just a platform for buying books, but a bridge connecting readers to the timeless values of culture and wisdom. With an elegant, user-friendly interface and an intelligent search system, we are committed to providing a quick and convenient shopping experience. Additionally, our special promotions and home delivery services ensure that you save time and fully enjoy the joy of reading. Let us accompany you on the journey of exploring knowledge and personal growth! textbookfull.com