Automated Image Segmentation and Asymmetry Analysis for Breast Using Infrared Images

This paper proposes an automatic approach to segmentation and asymmetry analysis for breast in infrared images. Hough transform, canny edge detection operator and other technologies are used to extract four feature curves that can uniquely separate the left and right breasts. These feature curves include the two parabolic curves describing the lower boundaries of the breasts, and the left and right body boundary curves. On the basis of segmentation, unsupervised learning technique, which is based on the k-mean clustering algorithm, is applied to classify each segmented pixel into certain number clusters. Asymmetric abnormalities can then be easily identified based on the pixel distribution. Experiments show that this approach is effectual and feasible and it has been of great practical value in the diagnosing the asymmetric abnormalities for breast using infrared images.