Table of Contents

Cover

Title Page

Dedication

Editor Biographies

List of Contributors

Foreword

Preface

About the Companion website

Chapter 1: Multilevel Image Segmentation Using Modified Genetic Algorithm (MfGA)-based Fuzzy C-Means

1.1 Introduction
1.2 Fuzzy C-Means Algorithm
1.3 Modified Genetic Algorithms
1.4 Quality Evaluation Metrics for Image Segmentation
1.5 MfGA-Based FCM Algorithm
1.6 Experimental Results and Discussion
1.7 Conclusion
References

Chapter 2: Character Recognition Using Entropy-Based Fuzzy C-Means Clustering

2.1 Introduction
2.2 Tools and Techniques Used
2.3 Methodology
2.4 Results and Discussion
2.5 Conclusion and Future Scope of Work
References
Appendix

Chapter 3: A Two-Stage Approach to Handwritten Indic Script Identification

3.1 Introduction
3.2 Review of Related Work
3.3 Properties of Scripts Used in the Present Work
3.4 Proposed Work
3.5 Experimental Results and Discussion
3.6 Conclusion
Acknowledgments
References

Chapter 4: Feature Extraction and Segmentation Techniques in a Static Hand Gesture Recognition System

4.1 Introduction
4.2 Segmentation Techniques
4.3 Feature Extraction Techniques
4.4 State of the Art of Static Hand Gesture Recognition Techniques
4.5 Results and Discussion
4.6 Conclusion
Acknowledgment
References

Chapter 5: SVM Combination for an Enhanced Prediction of Writers' Soft Biometrics

5.1 Introduction
5.2 Soft Biometrics and Handwriting Over Time
5.3 Soft Biometrics Prediction System
5.4 Experimental Evaluation
5.5 Discussion and Performance Comparison
5.6 Conclusion
References

Chapter 6: Brain-Inspired Machine Intelligence for Image Analysis: Convolutional Neural Networks

6.1 Introduction
6.2 Convolutional Neural Networks
6.3 Toward Understanding the Brain, CNNs, and Images
6.4 Conclusion
References

Chapter 7: Human Behavioral Analysis Using Evolutionary Algorithms and Deep Learning

7.1 Introduction
7.2 Human Action Recognition Using Evolutionary Algorithms and Deep Learning
7.3 Experimental Study
7.4 Conclusions and Future Work
References

Chapter 8: Feature-Based Robust Description and Monocular Detection: An Application to Vehicle tracking

8.1 Introduction
8.2 Extraction of Local Features by SIFT and SURF
8.3 Global Features: Real-Time Detection and Vehicle Tracking
8.4 Vehicle Detection and Validation
8.5 Experimental Study
8.6 Conclusions
References

Chapter 9: A GIS Anchored Technique for Social Utility Hotspot Detection

9.1 Introduction
9.2 The Technique
9.3 Case Study
9.4 Implementation and Results
9.5 Analysis and Comparisons
9.6 Conclusions
Acknowledgments
References

Chapter 10: Hyperspectral Data Processing: Spectral Unmixing, Classification, and Target Identification

10.1 Introduction
10.2 Background and Hyperspectral Imaging System
10.3 Overview of Hyperspectral Image Processing
10.4 Spectral Unmixing
10.5 Classification
10.6 Target Detection
10.7 Conclusions
References

Chapter 11: A Hybrid Approach for Band Selection of Hyperspectral Images

11.1 Introduction
11.2 Relevant Concept Revisit
11.3 Proposed Algorithm
11.4 Experiment and Result
11.5 Conclusion
References

Chapter 12: Uncertainty-Based Clustering Algorithms for Medical Image Analysis

12.1 Introduction
12.2 Uncertainty-Based Clustering Algorithms
12.3 Image Processing
12.4 Medical Image Analysis with Uncertainty-Based Clustering Algorithms
12.5 Conclusions
References

Chapter 13: An Optimized Breast Cancer Diagnosis System Using a Cuckoo Search Algorithm and Support Vector Machine Classifier

13.1 Introduction
13.2 Technical Background
13.3 Proposed Breast Cancer Diagnosis System
13.4 Results and Discussions
13.5 Conclusion
13.6 Future Work
References

Chapter 14: Analysis of Hand Vein Images Using Hybrid Techniques

14.1 Introduction
14.2 Analysis of Vein Images in the Spatial Domain
14.3 Analysis of Vein Images in the Frequency Domain
14.4 Comparative Analysis of Spatial and Frequency Domain Systems
14.5 Conclusion
References

Chapter 15: Identification of Abnormal Masses in Digital Mammogram Using Statistical Decision Making

15.1 Introduction
15.2 Previous Works
15.3 Proposed Method
15.4 Experimental Result
15.5 Result Evaluation
15.6 Comparative Analysis
15.7 Conclusion
Acknowledgments
References

Chapter 16: Automatic Detection of Coronary Artery Stenosis Using Bayesian Classification and Gaussian Filters Based on Differential Evolution

16.1 Introduction
16.2 Background
16.3 Proposed Method
16.4 Computational Experiments
16.5 Concluding Remarks
Acknowledgment
References

Chapter 17: Evaluating the Efficacy of Multi-resolution Texture Features for Prediction of Breast Density Using Mammographic Images

17.1 Introduction
17.2 Materials and Methods
17.3 Results
17.4 Conclusion and Future Scope
References

Index

End User License Agreement

List of Illustrations

Chapter 1: Multilevel Image Segmentation Using Modified Genetic Algorithm (MfGA)-based Fuzzy C-Means

Figure 1.1 Flowchart of MfGA-based FCM algorithm.
Figure 1.2 8-class segmented 256×256 grayscale Lena image with the class levels obtained by (a–d) FCM, (e–h) GA-based FCM, and (i–l) MfGA-based FCM algorithms of four results of Table 1.1, with $c01-math-128$ as the quality measure.
Figure 1.3 8-class segmented 256×256 grayscale Lena image with the class levels obtained by (a–d) FCM, (e–h) GA-based FCM, and (i–l) MfGA-based FCM algorithms of four results of Table 1.2, with $c01-math-129$ as the quality measure.
Figure 1.4 8-class segmented 256×256 grayscale peppers image with the class levels obtained by (a–d) FCM, (e–h) GA-based FCM, and (i–l) MfGA-based FCM algorithms of four results of Table 1.4, with $c01-math-134$ as the quality measure.
Figure 1.5 8-class segmented 256×256 grayscale peppers image with the class levels obtained by (a–d) FCM, (e–h) GA-based FCM, and (i–l) MfGA-based FCM algorithms of four results of Table 1.5, with $c01-math-135$ as the quality measure.
Figure 1.6 8-class segmented 256×256 grayscale baboon image with the class levels obtained by (a–d) FCM, (e–h) GA-based FCM, and (i–l) MfGA-based FCM algorithms of four results of Table 1.7, with $c01-math-140$ as the quality measure.
Figure 1.7 8-class segmented 256×256 grayscale baboon image with the class levels obtained by (a–d) FCM, (e–h) GA-based FCM, and (i–l) MfGA-based FCM algorithms of four results of Table 1.8, with $c01-math-141$ as the quality measure.

Chapter 2: Character Recognition Using Entropy-Based Fuzzy C-Means Clustering

Figure 2.1 Letters used for training the algorithm: (a) Arial, (b) Bardely, (c) Calibri, (d) Cambria, and (e) Times New Roman.
Figure 2.2 (a) Raw image, (b) converted binary image, and (c) boundary of the image.
Figure 2.3 Flow diagram of preprocessing technique.
Figure 2.4 Longest run feature row-wise and column-wise for a $c02-math-054$ image.
Figure 2.5 Flowchart of the proposed methodology for classification and recognition.
Figure 2.6 Visualization of cluster centers of input data obtained using the (a) FCM algorithm, (b) EFC algorithm, and (c) EFCM algorithm.

Chapter 3: A Two-Stage Approach to Handwritten Indic Script Identification

Figure 3.1 Segments of sample document pages written in: (a) Bangla, (b) Devanagari, (c) Gurumukhi, (d) Oriya, (e) Malayalam, (f) Telugu, (g) Urdu, and (h) Roman scripts, respectively.
Figure 3.2 Wavelet decomposition tree for a 2D image using both low-pass and high-pass filters.
Figure 3.3 Pictorial description of the LL, HL, LH, and HH components after applying Haar wavelet transform to the original grayscale word image written in Bangla script.
Figure 3.4 Illustration of a single projection at a specified rotation angle $c03-math-032$ on a handwritten Bangla word image.
Figure 3.5 Horizontal and vertical projections of a word image written in Bangla script.
Figure 3.6 Pictorial description of the geometry of the radon transformation.
Figure 3.7 Illustration of RT of the word images written in (a) Bangla, (b) Devanagari, (c) Gurumukhi, (d) Oriya, (e) Malayalam, (f) Telugu, (g) Urdu, and (h) Roman scripts, respectively.
Figure 3.8 Graph for: (a) normal distribution, (b) positively skewed distribution, and (c) negatively skewed distribution.
Figure 3.9 Illustration of (a) mesokurtic, (b) leptourtic, and (c) platykurtic curves.
Figure 3.10 Pictorial representation of the two-stage approach of the proposed script identification technique.
Figure 3.11 Sample word images of eight handwritten scripts from our database written in: (a) Bangla, (b) Devanagari, (c) Gurumukhi, (d) Oriya, (e) Malayalam, (f) Telugu, (g) Urdu, and (h) Roman, respectively.
Figure 3.12 Graphical comparison of model building time (seconds) required by seven different classifiers.
Figure 3.13 Comparison of multiple classifiers for: (a) Nemenyi's test and (b) Bonferroni–Dunn's test.
Figure 3.14 Confusion matrix for a classification rule.
Figure 3.15 Graph showing the performance of SVM classifier on the ROC curve for eight handwritten scripts.
Figure 3.16 Samples of successfully classified handwritten word images written in: (a) Bangla, (b) Devanagari, (c) Gurumukhi, (d) Oriya, (e) Malayalam, (f) Telugu, (g) Urdu, and (h) Roman scripts, respectively.
Figure 3.17 Sample handwritten word images misclassified by the present technique due to the presence of: (a) a significantly smaller number of characters constituting the word; (b) skewness; (c–e) structural similarity in Devanagari (misclassified as Gurumukhi), Gurumukhi (misclassified as Devanagari), and Oriya (misclassified as Bangla); (f) Matra-like structure in Roman script; and (g–h) abrupt spaces in Malayalam and Telugu scripts, respectively.

Chapter 4: Feature Extraction and Segmentation Techniques in a Static Hand Gesture Recognition System

Figure 4.1 Operational flowchart of proposed static hand gesture recognition system.
Figure 4.2 Flowchart of hand region extraction method.
Figure 4.3 Otsu method for thresholding:(a) gray-level image, (b) bimodal histogram, and (c) segmented binary image.
Figure 4.4 Hand gesture images captured in different angles.
Figure 4.5 Block diagram of homomorphic filtering.
Figure 4.6 (a) RGB color image, (b) segmented image, and (c) ROI after morphological operation.
Figure 4.7 Basic steps of feature extraction.
Figure 4.8 Zoning topology.
Figure 4.9 Uniform Background Database: (a) digit 1, (b) digit 2, and (c) digit 3.
Figure 4.10 Complex Background Database: (a) digit 0, (b) digit 1, and (c) digit 2.
Figure 4.11 (a) Input gesture, (b) YCbCr segmented image, and (c) extracted hand region after morphological operation.
Figure 4.12 (a) input gesture, (b) segmented image by our proposed method, and (c) extracted hand region after morphological operation.
Figure 4.13 ASL similar-shape gestures: (a) 7, (b) 8, and (c) 9.

Chapter 5: SVM Combination for an Enhanced Prediction of Writers' Soft Biometrics

Figure 5.1 Soft-biometrics prediction system.
Figure 5.2 HOG feature calculated on a handwritten text image.
Figure 5.3 GLBP feature extraction for each pixel.
Figure 5.4 IAM data set samples.
Figure 5.5 KHATT data set samples.
Figure 5.6 LBP operators; performances for gender prediction on the IAM-1 corpus.
Figure 5.7 Influence of the grid size for gender prediction on the IAM-1 corpus.

Chapter 6: Brain-Inspired Machine Intelligence for Image Analysis: Convolutional Neural Networks

Figure 6.1 A neuron.
Figure 6.2 Hidden layers.
Figure 6.3 A perceptron.
Figure 6.4 Gradient of a single weight.
Figure 6.5 Example of applying convolution to images.
Figure 6.6 Convolution kernels.
Figure 6.7 Edge detection in images.
Figure 6.8 Sobel operator.
Figure 6.9 Convolutional neural network: architecture.
Figure 6.10 Connections in a regular neural network versus a convolutional neural network.
Figure 6.11 Example of parameter sharing among neurons.
Figure 6.12 Example of pooling.
Figure 6.13 Input image.
Figure 6.14 Output from the first sublayer of the first layer.
Figure 6.15 Output from the first layer.
Figure 6.16 Output from the second layer.
Figure 6.18 Output from the fourth layer.
Figure 6.19 Output from the fifth layer.

Chapter 7: Human Behavioral Analysis Using Evolutionary Algorithms and Deep Learning

Figure 7.1 Pictorial representation of various steps in the proposed classification system. Best viewed in color.
Figure 7.2 Action bank features of observations in KTH: (i–iii) are videos of boxing action and (iv–vi) are videos of running.
Figure 7.3 CNN classifier for recognizing actions in videos.
Figure 7.4 Analysis of classification errors of solutions generated by the proposed hybrid training approach.
Figure 7.5 Visualization of EA population generated by the proposed approach for Set-1 of UCF50: (i) after initialization by EA and (ii) after training the solutions with BPA for $c07-math-031$ epochs.
Figure 7.6 Visualization of EA population generated by the proposed approach for Set-2 to Set-5 of UCF50. The sub-Figure (i), (ii) correspond to Set-2; (iii), (iv) are for Set-3; (v), (vi) correspond to Set-4 and (vii), (viii) are for Set-5.
Figure 7.7 Visualization of EA population generated for KTH: (i) visualization after initialization by EA and (ii) visualization after training with BPA.
Figure 7.8 Visualization of chromosomes (candidate solutions) generated by the proposed hybrid training approach for UCF50: (a) The pictorial representation of convolution masks and seed values corresponding to a chromosome X of size 64; (b) chromosomes generated for the KTH data set; and (c–g) correspond to chromosomes generated for UCF50 Set-1 to Set-5, respectively.

Chapter 8: Feature-Based Robust Description and Monocular Detection: An Application to Vehicle tracking

Figure 8.1 Illustration of Lowe's matching method.
Figure 8.2 Generic points aided robust description (GPRD) matching scheme.
Figure 8.3 Matching with Lowe's SIFT description.
Figure 8.4 Matching with GPRD descriptors.
Figure 8.5 Visual comparison of GPRD and Lowe's SIFT description: The frame on the left side (blue colored) is the result of GPRD, and the frame on the right side (red colored) is the result of the SIFT description.
Figure 8.6 Assessment of Haar-like features in the context of face recognition.
Figure 8.7 Recognition and tracking scheme.
Figure 8.8 Scan for lines' vertical response (right), and discontinuity on lines (left).
Figure 8.9 Screenshots of the recognition and tracking results on the Istanbul TEM highway.
Figure 8.10 The responses of true matchings for SIFT, SURF, and SIFT GPRD–based detection.
Figure 8.11 The responses of false matchings.
Figure 8.12 Evaluation of SIFT GPRD in a road traffic video.
Figure 8.13 Screenshots of recognition and tracking system result for TEM highway during rush hour.
Figure 8.14 Screenshots of recognition and tracking system results for LISA-Q Front FOV 1 during rush hour.
Figure 8.15 Comparison of true positive per frame metrics for global and local features.
Figure 8.16 Comparison of false positive per frame metrics for global and local features.

Chapter 9: A GIS Anchored Technique for Social Utility Hotspot Detection

Figure 9.1 Map of Egra police station, West Bengal, India.
Figure 9.2 Existing police stations of Egra.
Figure 9.3 Cluster formation after acquiring ranks.
Figure 9.4 Rank indicator.
Figure 9.5 Hotspot zone formation after considering Condition(1) and Condition(2).
Figure 9.6 Depicting the encircled “red” hotspot zone.
Figure 9.7 Splitting the “red” zone into two clusters, $c09-math-070$ and $c09-math-071$ .
Figure 9.8 Depicting suitable locations for construction of beat houses.
Figure 9.9 Before.
Figure 9.10 After.
Figure 9.11 Choosing options.
Figure 9.12 Creating a new profile.
Figure 9.13 Opening an existing profile.
Figure 9.14 Digitization of a raster map.
Figure 9.15 Data association.
Figure 9.16 “Click Here” button.
Figure 9.17 Crime-Info window.
Figure 9.18 Number of cases received against respective crime types.
Figure 9.19 Buttons under “Police Station” label.
Figure 9.20 In the year 2011 (3 suggested police stations).
Figure 9.21 In the year 2012 (2 suggested police stations).
Figure 9.22 In the year 2013 (2 suggested police stations).

Chapter 10: Hyperspectral Data Processing: Spectral Unmixing, Classification, and Target Identification

Figure 10.1 Hyperspectral imaging concept.
Figure 10.2 Diagrammatic representation of the schematics for a commonly used hyperspectral imaging system.
Figure 10.3 General processing steps involved for hyperspectral data.
Figure 10.4 Hyperspectral unmixing processing chain.
Figure 10.5 Conceptual diagram of a simple linear mixture model geometry.
Figure 10.6 Nonlinear mixture model.
Figure 10.7 Double concentric sliding window.

Chapter 11: A Hybrid Approach for Band Selection of Hyperspectral Images

Figure 11.1 Electromagnetic spectrum [5].
Figure 11.2 AVIRIS hyperspectral image cube [6].
Figure 11.3 Basic biological immune system [33].
Figure 11.4 Flowchart of the clonal selection technique.
Figure 11.5 Flowchart of the proposed technique.
Figure 11.6 Indiana Pines image. (a) Indiana Pines image. (b) Ground truth.
Figure 11.7 Pavia image. (a) Three-band color composite image. (b) Ground truth.
Figure 11.8 Overall accuracy of the proposed algorithm on the Indiana data set.
Figure 11.9 Comparison of the proposed method with MI, WaLuDi, and TMI in terms of overall accuracy for the Indiana data set.
Figure 11.10 Comparison of the proposed method with MI, WaLuDi, and TMI in terms of overall accuracy for the Pavia Center data set.
Figure 11.11 Basics of a SVM.
Figure 11.12 Comparison of the proposed method (2D PCA and fuzzy KNN) with 2D PCA and SVM in terms of overall accuracy for the Indiana data set.
Figure 11.13 Comparison of the proposed method (2D PCA) and LDA in terms of overall accuracy for the Indiana data set.

Chapter 12: Uncertainty-Based Clustering Algorithms for Medical Image Analysis

Figure 12.1 MRI image segmentation using (a) FCM, (b) $c012-math-050$ , and (c) $c012-math-051$ .
Figure 12.2 Segmentation results on (a) original image, (b) same image with mixed noise, results of (c) FCM_S1, (d) FCM_S2, (e) EnFCM, (f) FGFCM_S1, (g) FGFCM_S2, and (h) FGFCM.
Figure 12.3 MRI image – speckle noise.
Figure 12.4 Noisy image segmentation. (a) FCM, (b) sFCM1,1, (c) $c012-math-054$ , (d) $c012-math-055$ , (e) $c012-math-056$ , (f) sIFCM $c012-math-057$ , and (g) $c012-math-058$ .
Figure 12.5 (a) Original image. Segmented images of leukemia using (b) FCM, (c) $c012-math-098$ , (d) $c012-math-099$ , and (e) $c012-math-100$ .
Figure 12.6 (a) Original image. Segmented images of leukemia using (b) FCM, (c) IFCM, (d) $c012-math-101$ , (e) $c012-math-102$ , and (f) $c012-math-103$ .

Chapter 13: An Optimized Breast Cancer Diagnosis System Using a Cuckoo Search Algorithm and Support Vector Machine Classifier

Figure 13.1 Proposed breast cancer diagnosis system.
Figure 13.2 Sample mammographic images.
Figure 13.3 Enhanced mammographic images.
Figure 13.4 Otsu's thresholded and morphological segmented breast tissues.

Chapter 14: Analysis of Hand Vein Images Using Hybrid Techniques

Figure 14.1 The basic block diagram of a multimodal biometric recognition system.
Figure 14.2 Flow diagram of the analysis of vein images in the spatial domain.
Figure 14.3 Input hand vein images (first row). Preprocessed vein images (second row).
Figure 14.4 Output of Gabor filter for Palm vein images with $c014-math-060$ : (a) $c014-math-061$ , $c014-math-062$ ; (b) $c014-math-063$ , $c014-math-064$ ; (c) $c014-math-065$ , $c014-math-066$ ; (d) $c014-math-067$ , $c014-math-068$ ; (e) $c014-math-069$ , $c014-math-070$ ; (f) $c014-math-071$ , $c014-math-072$ ; (g) $c014-math-073$ , $c014-math-074$ ; (h) $c014-math-075$ , $c014-math-076$ ; and (i) $c014-math-077$ , $c014-math-078$ .
Figure 14.6 Output of Gabor filter for Wrist vein images with $c014-math-079$ : (a) $c014-math-080$ , $c014-math-081$ ; (b) $c014-math-082$ , $c014-math-083$ ; (c) $c014-math-084$ , $c014-math-085$ ; (d) $c014-math-086$ , $c014-math-087$ ; (e) $c014-math-088$ , $c014-math-089$ ; (f) $c014-math-090$ , $c014-math-091$ ; (g) $c014-math-092$ , $c014-math-093$ ; (h) $c014-math-094$ , $c014-math-095$ ; and (i) $c014-math-096$ , $c014-math-097$ .
Figure 14.7 Output of Gabor filter for Finger vein images with $c014-math-098$ : (a) $c014-math-099$ , $c014-math-100$ ; (b) $c014-math-101$ , $c014-math-102$ ; (c) $c014-math-103$ , $c014-math-104$ ; (d) $c014-math-105$ , $c014-math-106$ ; (e) $c014-math-107$ , $c014-math-108$ ; (f) $c014-math-109$ , $c014-math-110$ ; (g) $c014-math-111$ , $c014-math-112$ ; (h) $c014-math-113$ , $c014-math-114$ ; and (i) $c014-math-115$ , $c014-math-116$ .
Figure 14.8 Fusion of two modalities of hand vein features.
Figure 14.9 Fusion of three modalities of hand vein features.
Figure 14.10 Fusion of all four modalities of hand vein features.
Figure 14.11 Flow diagram of the analysis of vein images in the frequency domain.
Figure 14.12 Fusion for the hand vein images in the frequency domain.
Figure 14.13 A linear support vector machine.
Figure 14.14 Preprocessed hand vein images at various stages.
Figure 14.15 Contourlet-transformed vein images: (a) dorsal hand vein, (b) palm vein, (c) wrist vein, and (d) finger vein.

Chapter 15: Identification of Abnormal Masses in Digital Mammogram Using Statistical Decision Making

Figure 15.1 Schematic diagram of the proposed method.
Figure 15.2 (a) The original mammogram image and (b) the prepared mammogram image (MIAS mdb184.L). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society. [31]
Figure 15.3 Full and complete binary tree.
Figure 15.4 Enhanced mammogram image (MIAS mdb184.L). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.5 (a) Edge map of Level 1, (b) edge map of Level 2, and (c) edge map of Level 3 (MIAS mdb184.L). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.6 (a) The edge map of a mammogram, (b) showing the layers of pectoral muscle, and (c) showing inverted triangles marked by different gray shades (MIAS mdb184.L). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.7 Pectoral muscle lies within gray-shaded derived rectangular area (MIAS mdb184.L). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.8 Isolated pectoral boundary (MIAS mdb184.L). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.9 The technique of traversing process. (a) Black cell represents the current pixel, gray cell is representative of already traversed pixel, and the rest are the path for further traversing. (b) The priority of selection of neighbor is clockwise. Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.10 Detected breast contour (MIAS mdb184.L). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.11 (a) Breast ROI and (b) Boundaries of anatomical regions within breast ROI (MIAS mdb184.L). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.12 Intensity distribution of regions after coloring (MIAS mdb184.L). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.13 (a) Highlighted regions with abnormal masses and (b) boundary of abnormal regions (MIAS mdb184.L). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.14 MIAS mdb272.L: (a) mammogram image, (b) segmented anatomical regions without highlighted abnormality, and (c) derived image showing absence of boundary of abnormal region(s). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.15 MIAS mdb028.L: (a) mammogram image, (b) segmented anatomical regions with highlighted abnormality, and (c) derived image showing boundary of abnormal region(s). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.16 MIAS mdb001.R: (a) mammogram image, (b) segmented anatomical regions with highlighted abnormality, and (c) derived image showing boundary of abnormal region(s). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.17 MIAS mdb145.R: (a) mammogram image, (b) segmented anatomical regions with highlighted abnormality, and (c) derived image showing boundary of abnormal region(s). Source: Suckling (1994) [31]. Reproduced with permission of Mammographic Image Analysis Society.
Figure 15.18 The Z score analysis graph for mammogram mdb272.L.
Figure 15.19 score analysis graph for mammogram mdb028.L.
Figure 15.20 The Z score analysis graph for mammogram mdb001.R.
Figure 15.21 The Z score analysis graph for mammogram mdb145.R.
Figure 15.22 Empirical ROC curve for tumor identification.

Chapter 16: Automatic Detection of Coronary Artery Stenosis Using Bayesian Classification and Gaussian Filters Based on Differential Evolution

Figure 16.1 Two X-ray angiograms with detection of coronary stenosis performed by a cardiologist.
Figure 16.2 (a) X-ray coronary angiogram. (b) Ground-truth image of angiogram in (a). (c) Gaussian profile of the method of Chaudhuri et al. [16]. Second row: Template using $c016-math-016$ , $c016-math-017$ , $c016-math-018$ , with $c016-math-019$ , and $c016-math-020$ , respectively, and resulting filtered image in (f). Last row: Matching template using $c016-math-021$ , $c016-math-022$ , $c016-math-023$ , with $c016-math-024$ , and $c016-math-025$ , respectively, and resulting filtered image in (i).
Figure 16.3 Ackley function in two dimensions. (a) Isometric view in $c016-math-068$ and $c016-math-069$ , and (b) level plot of the function, where the optimal value is located at $c016-math-070$ and $c016-math-071$ .
Figure 16.4 Numerical example for solving the 2D Ackley function using DE as an optimization strategy.
Figure 16.5 First row: Segmentation results using the Ridler and Calvard method. The remaining three rows illustrate the results of length filtering using 100, 200, and 500 pixels as connected components, respectively.
Figure 16.6 (a) X-ray coronary angiogram. (b) Skeleton of segmented vessel. (c) Addition of skeleton and boundary pixels. (d) Skeleton using normalized intensities as Euclidean distance. (e) Separation of vessel segments using bifurcation pixels. (f, g) Detection of local minima points over Gaussian filter response and original angiogram, respectively. (h) Stenosis detection marked in a black circle by cardiologist.
Figure 16.7 (a) Stenosis pattern of $c016-math-154$ pixels. (b) Histogram of vessel width estimation of pattern in (a). (c) No stenosis pattern of $c016-math-155$ pixels. (d) Histogram of vessel width estimation of pattern in (c).
Figure 16.8 First row: Subset of X-ray angiograms. Second row: Ground-truth images. The remaining six rows present the Gaussian filter response of the methods of Kang et al. [25], Al-Rawi et al. [21], Cruz et al. [27], Chaudhuri et al. [16], and Cinsdikici et al. [20], and the proposed method, respectively.
Figure 16.9 First row: Subset of X-ray angiograms. Second row: Ground-truth images. The remaining five rows present the segmentation results of the methods of Kapur et al. [33], histogram concavity [34], Pal and Pal [35], RATS [36], and Ridler and Calvard [32], respectively.
Figure 16.10 First column: Subset of X-ray angiograms. Second column: Ground-truth images. Third column: Segmentation result obtained from the proposed method. Last column: Product between segmentation result and input angiogram.
Figure 16.11 First row: Subset of patterns of no-stenosis cases. Second column: Subset of vessel stenosis patterns.

Chapter 17: Evaluating the Efficacy of Multi-resolution Texture Features for Prediction of Breast Density Using Mammographic Images

Figure 17.1 CAD system for breast density classification. Note: Shaded blocks indicate the steps involved in present work.
Figure 17.2 MIAS database breast density classification.
Figure 17.3 Sample mammograms showing typical cases of breast tissue density: (a) typical F tissue $c017-math-001$ , (b) typical FG tissue $c017-math-002$ , and (c) typical DG tissue $c017-math-003$ .
Figure 17.4 Sample mammograms showing atypical cases of breast tissue density: (a) atypical F tissue $c017-math-004$ , (b) atypical FG tissue $c017-math-005$ , and (c) atypical DG tissue $c017-math-006$ .
Figure 17.5 Database description.
Figure 17.6 ROI extraction protocol $c017-math-009$ .
Figure 17.7 Sample ROIs: (a) typical F ROI $c017-math-010$ ; (b) typical FG ROI $c017-math-011$ ; (c) typical DG ROI $c017-math-012$ ; (d) atypical F ROI $c017-math-013$ ; (e) atypical FG ROI $c017-math-014$ ; and (f) atypical DG ROI $c017-math-015$ .
Figure 17.8 Block diagram: workflow for prediction of breast density.
Figure 17.9 Different feature extraction techniques used in texture analysis. Note: GLCM: gray-level co-occurence matrix; GLRLM: gray-level run length matrix; NGTDM: neighborhood gray tone difference matrix; SFM: statistical feature matrix; FPS: Fourier power spectrum; STFT: short-time Fourier transform; 2D-DWT: two-dimensional discrete wavelet transform; WPT: wavelet packet transform; NSCT: non-subsampled countourlet transform; NSST: non-subsampled shearlet transform.
Figure 17.10 Time-domain, frequency-domain, STFT, and wavelet analysis of a signal.
Figure 17.11 Process of wavelet analysis of an image.
Figure 17.12 Wavelet transform of sample Lena image.
Figure 17.13 Wavelet transform of image F mdb132.
Figure 17.14 Wavelet decomposition of an image up to the second level.
Figure 17.15 (a) 2D wavelet decomposition of image up to the second level. (b) 2D wavelet decomposition of sample image $c017-math-018$ using a Haar wavelet filter up to the second level.
Figure 17.16 (a) Haar wavelet function $c017-math-039$ and (b) Haar scaling function $c017-math-040$ .
Figure 17.17 SVM classifier for linearly separable data.
Figure 17.18 Example to illustrate the SVM algorithm.
Figure 17.19 SVM for nonlinearly separable data.
Figure 17.20 Flowchart of proposed CAD system design.
Figure 17.21 Proposed CAD system design.

List of Tables

Chapter 1: Multilevel Image Segmentation Using Modified Genetic Algorithm (MfGA)-based Fuzzy C-Means

Table 1.1 Class boundaries and evaluated segmentation quality measures $c01-math-103$ by different algorithms for different classes of the Lena image
Table 1.2 Class boundaries and evaluated segmentation quality measures $c01-math-104$ by different algorithms for different classes of the Lena image
Table 1.3 Different algorithm-based means and standard deviations using different types of fitness functions and mean of time taken by different algorithms for the Lena image
Table 1.4 Class boundaries and evaluated segmentation quality measures $c01-math-114$ by different algorithms for different classes of the peppers image
Table 1.5 Class boundaries and evaluated segmentation quality measures $c01-math-115$ by different algorithms for different classes of the peppers image
Table 1.6 Different algorithm-based mean and standard deviation using different types of fitness functions and mean of time taken by different algorithms for the peppers image
Table 1.7 Class boundaries and evaluated segmentation quality measures $c01-math-120$ by different algorithms for different classes of the baboon image
Table 1.8 Class boundaries and evaluated segmentation quality measures $c01-math-121$ by different algorithms for different classes of the baboon image
Table 1.9 Different algorithm-based mean and standard deviation using different types of fitness functions and mean of time taken by different algorithms for the baboon image

Chapter 2: Character Recognition Using Entropy-Based Fuzzy C-Means Clustering

Table 2.1 Output of different clustering algorithms for input data
Table 2.2 Results of recognition of alphabets with three algorithms for Times New Roman
Table 2.3 Recognition accuracy with respect to each font

Chapter 3: A Two-Stage Approach to Handwritten Indic Script Identification

Table 3.1 Important information related to scripts [26] used in the present work
Table 3.2 Success rates of the proposed script identification technique using seven well-known classifiers
Table 3.3 Recognition accuracies of seven classifiers and their corresponding ranks on 12 different data sets (ranks in parentheses are used for performing the Friedman test)
Table 3.4 Statistical performance measures along with their respective means (shaded in gray) achieved by the proposed technique for eight handwritten scripts
Table 3.5 Comparison of statistical performance parameters for the four cases (best case is styled in bold)
Table 3.6 Comparison of the present script identification result with state-of-the art methods

Chapter 4: Feature Extraction and Segmentation Techniques in a Static Hand Gesture Recognition System

Table 4.1 User-dependent and user-independent classification results
Table 4.2 Confusion matrix of the Krawtchouk moment for user-independent condition
Table 4.3 Confusion matrix of the Tchebichef moment for the user-independent condition
Table 4.4 Confusion matrix of the geometric moment for the user-independent condition
Table 4.5 Classification result of Krawtchouk moment zonal features
Table 4.6 Classification results of serial, parallel, and MLE-based hidden feature fusion
Table 4.7 Classification result of F-ratio-based enhanced features

Chapter 5: SVM Combination for an Enhanced Prediction of Writers' Soft Biometrics

Table 5.1 Influence of SVM kernels for gender prediction on the IAM-1 corpus (%)
Table 5.2 Results of individual systems for gender prediction (%)
Table 5.3 Results of combination systems for gender prediction (%)
Table 5.4 Results of handedness prediction for individual systems (%)
Table 5.5 Results of combination systems for handedness prediction (%)
Table 5.6 Results of handedness prediction for individual systems (%)
Table 5.7 Results of combination systems for age prediction (%)
Table 5.8 State-of-the-art results

Chapter 6: Brain-Inspired Machine Intelligence for Image Analysis: Convolutional Neural Networks

Table 6.1 Conceptual differences between CNN and the brain/visual system
Table 6.2 Activation functions
Table 6.3 Factors and assigned variables for the perceptron model

Chapter 7: Human Behavioral Analysis Using Evolutionary Algorithms and Deep Learning

Table 7.1 The variation in fitness value of EA populations over iterations for UCF50
Table 7.2 Accuracy (in %) of candidate solutions generated for UCF50 using neural network classifier
Table 7.3 Performance (in %) of candidate solutions generated for UCF50 using ELM classifier
Table 7.4 Accuracy (in # misclassified observations) using fusion on UCF50
Table 7.5 Confusion matrix for UCF50
Table 7.6 Performance on the UCF50 data set
Table 7.7 Performance (in %) of CNN features with NN and ELM classifiers generated by the proposed approach [with back-propagation algorithm (BPA) and evolutionary algorithms (EA)] on the KTH data set
Table 7.8 Performance (in %) on the KTH data set
Table 7.9 Accuracy (in %) of CNN classifiers using BPAs, EAs and the hybrid approach for UCF50
Table 7.10 Accuracy (in %) of the proposed approach using NN and ELM classifiers for UCF50

Chapter 8: Feature-Based Robust Description and Monocular Detection: An Application to Vehicle tracking

Table 8.1 Performance results
Table 8.2 Performance results (video data set belonging to [12])

Chapter 9: A GIS Anchored Technique for Social Utility Hotspot Detection

Table 9.1 Hotspot-detecting tasks, depending on respective factors
Table 9.2 Table depicting the least, moderate, and highest crime-prone regions situated in seven police stations throughout West Bengal State, India
Table 9.3 Classification of crime types with respect to their rankings
Table 9.4 Table depicting the numerator and denominator values of respective regions calculated with respect to the crime types (as obtained from Table 9.3)
Table 9.5 Table delineating the final values of $c09-math-040$ 's and $c09-math-041$ 's (data for $c09-math-042$ =1)
Table 9.6 Table delineating the final values of $c09-math-043$ 's and $c09-math-044$ 's (data for $c09-math-045$ =5)
Table 9.7 Table delineating the final values of $c09-math-046$ 's and $c09-math-047$ 's (data for $c09-math-048$ =10)
Table 9.8 Values of the unknown variables used in Equation (9.1)
Table 9.9 The ranks of respective Anchals (arranged in descending order of rank)
Table 9.10 Depicting the comparative study between K-means and the proposed method
Table 9.11 Portraying the comparative study between the fuzzy clustering method and proposed method
Table 9.12 Illustrating the comparative study between ISODATA and the proposed method
Table 9.13 The comparative study between STAC and proposed method is drawn out
Table 9.14 Illustrating the advantages of the proposed method over RADIUS methodology
Table 9.15 Comparison with MCE – a technique for hotspot detection related to landslides

Chapter 11: A Hybrid Approach for Band Selection of Hyperspectral Images

Table 11.1 Indian Pines data set: classes with number of samples
Table 11.2 Pavia data set: classes with number of samples
Table 11.3 Selected bands for Indian Pines data set obtained by the proposed method
Table 11.4 Accuracy of classification using the algorithm

Chapter 12: Uncertainty-Based Clustering Algorithms for Medical Image Analysis

Table 12.1 Cluster evaluation results on speckle noise image
Table 12.2 Performance indices of sFCM on leukemia image
Table 12.3 Performance indices of sIFCM on leukemia image

Chapter 13: An Optimized Breast Cancer Diagnosis System Using a Cuckoo Search Algorithm and Support Vector Machine Classifier

Table 13.1 Comparison of classification accuracy of proposed approach with the existing methods
Table 13.2 Comparison of computational costs

Chapter 14: Analysis of Hand Vein Images Using Hybrid Techniques

Table 14.1 Analysis of hand vein images in unimodal mode
Table 14.2 Analysis of hand vein images in multimodal mode
Table 14.3 Frequency domain analysis of hand vein images
Table 14.4 Comparative analysis of related work on vein-based biometric recognition

Chapter 15: Identification of Abnormal Masses in Digital Mammogram Using Statistical Decision Making

Table 15.1 Color lookup table
Table 15.2 Confusion matrix of response data reported from testing
Table 15.3 Observed operating points
Table 15.4 Accuracy measures based on size of mass detected by the proposed method
Table 15.5 Quantitative measures applied to assess the proposed methods
Table 15.6 Comparative analysis of proposed method with others

Chapter 16: Automatic Detection of Coronary Artery Stenosis Using Bayesian Classification and Gaussian Filters Based on Differential Evolution

Table 16.1 Training records for predicting cardiovascular risk
Table 16.2 Comparative analysis of $c016-math-158$ values with the testing set, using the proposed method and five GMF-based methods of the state of the art
Table 16.3 Comparative analysis of five automatic thresholding methods over the Gaussian filter response using the test set of X-ray angiograms
Table 16.4 Results of naive Bayes classifier over the test set of 20 records
Table 16.5 Confusion matrix for the test set of 20 records

Chapter 17: Evaluating the Efficacy of Multi-resolution Texture Features for Prediction of Breast Density Using Mammographic Images

Table 17.1 Description of studies carried out for classification of tissue density as fatty or dense on the MIAS database
Table 17.2 Description of studies carried out for classification of tissue density as fatty, fatty-glandular, dense-glandular, or extremely dense on the MIAS database
Table 17.3 Description of studies carried out for classification of tissue density as fatty, fatty-glandular, or dense-glandular on the MIAS database
Table 17.4 Properties of wavelet filters used
Table 17.5 Description of FDVs
Table 17.6 Experiment descriptions
Table 17.7 Classification performance of SVM classifier using different FDVs
Table 17.8 Classification performance of SSVM classifier with different FDVs
Table 17.9 Comparison of computational time for prediction of testing instances

Hybrid Intelligence for Image Analysis and Understanding

Edited by

Siddhartha Bhattacharyya

RCC Institute of Information Technology
India

Indrajit Pan

RCC Institute of Information Technology
India

Anirban Mukherjee

RCC Institute of Information Technology
India

Paramartha Dutta

Visva-Bharati University
India

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by law. Advice on how to obtain permission to reuse material from this title is available at http://www.wiley.com/go/permissions.

The right of Siddhartha Bhattacharyya, Indrajit Pan, Anirban Mukherjee, and Paramartha Dutta to be identified as the authors of the editorial material in this work has been asserted in accordance with law.

John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, UK

For details of our global editorial offices, customer services, and more information about Wiley products visit us at www.wiley.com.

Wiley also publishes its books in a variety of electronic formats and by print-on-demand. Some content that appears in standard print versions of this book may not be available in other formats.

While the publisher and authors have used their best efforts in preparing this work, they make no representations or warranties with respect to the accuracy or completeness of the contents of this work and specifically disclaim all warranties, including without limitation any implied warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by sales representatives, written sales materials or promotional statements for this work. The fact that an organization, website, or product is referred to in this work as a citation and/or potential source of further information does not mean that the publisher and authors endorse the information or services the organization, website, or product may provide or recommendations it may make. This work is sold with the understanding that the publisher is not engaged in rendering professional services. The advice and strategies contained herein may not be suitable for your situation. You should consult with a specialist where appropriate. Further, readers should be aware that websites listed in this work may have changed or disappeared between when this work was written and when it is read. Neither the publisher nor authors shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages.

Names: Bhattacharyya, Siddhartha, 1975- editor. | Pan, Indrajit, 1983- editor. | Mukherjee, Anirban, 1972- editor. | Dutta, Paramartha, editor.

Title: Hybrid intelligence for image analysis and understanding / edited by Siddhartha Bhattacharyya, Indrajit Pan, Anirban Mukherjee, Paramartha Dutta.

Classification: LCC TA1637 (ebook) | LCC TA1637 .H93 2017 (print) | DDC 621.36/7028563-dc23

Dedicated to my parents, the late Ajit Kumar Bhattacharyya and the late Hashi Bhattacharyya; my beloved wife, Rashni; my elder sisters, Tamali, Sheuli, and Barnali; my cousin sisters, Sutapa, Mousumi, and Soma; and all my students, who have made this journey enjoyable.