A Robust Graph Theoretic Approach for Image Segmentation

Abstract

This paper presents a new robust graph theoretic approach for image segmentation. The proposed method which is capable of accurately locating region boundaries has the following salient features. First, it is a nonsupervised approach which reflects the non-local properties of the image. Second, it guarantees that the regions are connected. Finally, it produces robust results which is almost unaffected by the influences of outliers. In this technique, at each step, a minimum weight edge is selected and the two regions connected by the minimum weight edge are considered for merge. The merging of regions is carried out, if the mean of the edges connecting the two regions is smaller than the maximum of the mean of the intra region edges along with the threshold value.

Pectoral muscle identification in mammograms

“…If there is an exact match between the ground truth and the algorithm identified pectoral muscle, then the value of FP and FN is zero. The mean of the error terms over "n" images can be calculated as: (20)…”

“…Hence, a novel merging algorithm, which is particularly more suitable for the pectoral muscle segmentation, is proposed to overcome this problem. The proposed merging algorithm is based on the work of Camilus et al (20) The proposed approach produced better results in terms of accuracy. We also demonstrate, in Section III below, the improved performance of the proposed approach by comparing it's accuracy with many related state-of-the-art methods.…”