EMPIRICAL EVALUATION OF LBP AND ITS DERIVATES FOR ABNORMALITY DETECTION IN MAMMOGRAM IMAGES

Digital image processing techniques are useful in abnormality detection in mammogram images. Recently, texture based image segmentation of mammogram images has become popular due to its better precision and accuracy. Local Binary Pattern has been a recently proposed texture descriptor which attracted the research community rigorously towards texture based analysis of digital images. Many texture descriptors have been developed as variants of Local Binary Pattern, because of its success. In this work, the performance of Local Binary Pattern descriptor and its variants namely Local Ternary pattern, Extended Local Ternary Pattern, Local Texture Pattern and Local Line Binary Pattern are evaluated for mammogram image segmentation using a supervised KNN algorithm. Performance metrics such as accuracy, error rate, sensitivity, specificity, Under Estimation Fraction and Over Estimation Fraction are used for comparison purpose. The results show that Local Binary Pattern outperforms other descriptors in terms of abnormality detection in mammogram images.