Classification of digital mammograms into masses and non-masses using texture combination features and SVM

Abstract

This paper presents a framework for grouping of mammogram images into carcinogenic and non-carcinogenic images. In this paper, image is enhanced using nonlinear contrast enhancement algorithm and then segmented using internal and external mask segmentation. The features used in the proposed work are based on the texture feature which means that the segmentation results will have minimal impact on the classification results of mammograms into masses and non – masses. The features used are Contrast, Energy, Homogeneity, Entropy, Mean Intensity, Standard Deviation, Taxonomic distinctness, Taxonomic distance and learning takes place by Support Vector Machine. The algorithm was applied on Mammographic Image Analysis Society database and when compared with contemporary techniques, the results were improved.