Texture Descriptor and Applications for Automated Diagnosis of Low Attenuating Lung Diseases

Texture Descriptor and Applications for Automated Diagnosis of Low Attenuating Lung Diseases
Issue Date
Texture represents the coarseness and statistical characteristics of the local variation of brightness between neighboring pixels, and plays an important role in image analysis and pattern recognition. Texture analysis is indispensable for important applications, such as in medical image analysis, document analysis, target detection, industrial surface detection, and remote sensing. Texture is usually represented by the texture feature which extracted using texture descriptor. A good texture descriptor must able to extract a stable and high discriminating texture feature. However, textures often vary due to variations in illumination, scaling changes, blurring effects, or other visual appearance perturbations. This will causes texture descriptor to generate variant texture feature which is hard to be classified. Although there are a lot of solutions to overcome the variance problems, most of the textures often exclude high discriminating properties. In this thesis, novel texture descriptors are proposed where some of the texture descriptors able to overcome those drawbacks with a high classification accuracy using degraded texture images. Two texture descriptors are constructed on spatial domain and one texture descriptor is constructed in frequency domain. The spatial domain- based texture descriptors are the Local Neighbor Differences (LND) and the Combined Neighborhood Differences (CND). LND uses the differences between each of the neighboring pixels which then are thresholded into 8-bit binary code while for CND, the local differences of surrounding neighborhood difference and centralized neighborhood difference pixels are compared and converted into binary 8-bit codeword. A binomial factor is assigned to the binary code in order to form a unique LND and CND values which represents the texture feature of the local neighborhood. The distribution of the LND and CND values is then computed to form a 256 dimensional histogram which represent the description of the
Chapter 1: Introduction 1 1.1 Introduction 1 1.2 Motivation 3 1.3 Contributions 6 1.4 Thesis Outline 7 Chapter 2: Backgrounds 8 2.1 Common Texture Descriptor 8 2.1.1 Gabor Filter 8 2.1.2 Local Binary Patterns 10 2.2 Computer Aided Diagnosis System for Lung Disease Analysis 11 2.2.1 Low Attenuating Lung Diseases 11 2.2.2 Automatic Diagnosis System 16 2.2.3 Feature Extraction 17 Chapter 3: Proposed Texture Descriptor 20 3.1 Local Neighbor Differences 20 3.2 Combined Neighborhood Differences 22 3.3 Local Circular Difference Phase Patterns 24 3.3.1 Main algorithm 25 3.3.2 The Properties 28 Chapter 4: Experimental Evaluation on Texture Descriptor 38 4.1 Experiment Setup 38 4.2 Texture Classification with Original Dataset 40 4.3 Invariance Evaluation 43 4.3.1 Invariance Performance 43 4.3.2 Texture Classification with Modified Dataset 43 Chapter 5: Proposed Automated Lung Disease Diagnosis 50 5.1 Background 50 5.2 Abnormal Region Pre-detection Method 51 5.3 Feature Extraction 54 Chapter 6: Evaluations of the Proposed Automated Lung Disease Diagnosis 58 6.1 Experiment Setup 50 6.2 Experimental Results and Discussions 51 6.2.1 Experiment on the Pre-detection Method 61 6.2.2 Evaluation of the Classification Performance 63 Chapter 7: Conclusion and Future work 77 7.1 Conclusions 77 7.2 Future Works 78 References 79 Acknowledgement 87
Appears in Collections:
College of Engineering(공과대학) > Electronic Engineering (전자공학) > Theses(전자공학 석박사 학위논문)
Files in This Item:

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.