An Unsupervised Segmentation Algorithm for Breast Ultrasound Images Using Local Histogram Features

Several segmentation methods have been presented for breast ultrasound (BUS) images. Unfortunately most of them are supervised and semi-automatic in nature. In this paper, a complete unsupervised algorithm for BUS image segmentation algorithm using local intensity and texture histograms features has been proposed. The texture and intensity features are combined in the clustering process. Initially the image is filtered using a texture preserving de-noising filter. A new texture feature is extracted from the filtered image. Using these features, employing a non-parametric Bayesian clustering method, image is segmented. This clustering is completely unsupervised in nature as no seeding or learning is required for this algorithm. Qualitative and quantitative segmentation results of images from BUS image databases prove the competitiveness of the proposed algorithm.