Different approaches for extracting information from the co-occurrence matrix.

Area of Science:

• Computer Vision
• Image Analysis
• Machine Learning

Background:

• Haralick's 1979 co-occurrence matrix method is a foundational technique for image texture analysis.
• Existing methods may not fully capture complex textural information present in images.

Purpose of the Study:

• To investigate novel texture descriptors derived from the co-occurrence matrix.
• To compare and combine different strategies for extending these texture descriptors.
• To enhance image classification performance in medical imaging.

Main Methods:

• Extraction of novel texture descriptors from multi-scale co-occurrence matrices and subwindows.
• Comparison of standard Haralick features with 3D shape, gray-level run-length, and principal component analysis (PCA) projection methods.
• Training of support vector machines (SVM) and ensemble models using the extracted texture descriptors.

Main Results:

• Novel texture descriptor extraction methods demonstrate improved performance over standard techniques.
• Validation across six diverse medical imaging datasets confirms the efficacy of the proposed approaches.
• The Wilcoxon signed rank test was used for statistical validation.

Conclusions:

• The developed novel texture descriptors offer a significant advancement in image texture analysis.
• These methods enhance the accuracy of image classification tasks, particularly in medical applications.
• The study provides a comprehensive comparison of various texture descriptor extraction strategies.