Classification of gray images based on their textural features is one of the main tools in medical image processing. Gray Level Co-occurrence Matrix (GLCM) is such a widely used technique which represents how frequently the different gray level combinations occur in an image, traditionally in Cartesian directions. Contrast, correlation, energy and homogeneity are features based on the calculated GLCM. However, in human anatomy, structures often take a curvilinear pattern and therefore the Cartesian GLCM may not be very efficient in medical imaging. In this study, an algorithm was developed to calculate the GLCM in radial and circumferential directions. The texture parameters calculated using the polar GLCM were then tested against those calculated using the traditional Cartesian GLCM, by means of simulated images with varying speckle features. Our results show that the Polar GLCM is better at detecting changes in number of speckles in radially and circumferentially oriented speckled images than the Cartesian GLCM even in the presence of noise.

Annual Research Symposium University of Colombo, 2018