Abstract. Medical image segmentation appears to be governed by the global intensity level and should be robust to local intensity fluctuation. We develop an efficient spectral graph method which seeks the best segmentation on a stack of gamma transformed versions of the original image. Each gamma image produces two types of grouping cues operating at different ranges: Short-range attraction pulls pixels towards region centers, while long-range repulsion pushes pixels away from region boundaries. With rough pixel correspondence between gamma images, we obtain an aligned cue stack for the original image. Our experimental results demonstrate that cutting across the entire gamma stack delivers more accurate segmentations than commonly used watershed algorithms. 1

Many applications need to segment out all small round regions in an image. This task of finding dots can be viewed as a region segmentation problem where the dots form one region and the areas between dots form the other. We formulate it as a graph cuts problem with two types of grouping cues: short-range attraction based on feature similarity and long-range repulsion based on feature dissimilarity. The feature we use is a pixel-centric relational representation that encodes local convexity: Pixels inside the dots and outside the dots become sinks and sources of the feature vector. Normalized cuts on both attraction and repulsion pop out all the dots in a single binary segmentation. Our experiments show that our method is more accurate and robust than state-of-art segmentation algorithms on four categories of microscopic images. It can also detect textons in natural scene images with the same set of parameters. a: b: microscopic images hand-labeled dots 1.