Hill-Climbing Based Image Segmentation

In this paper we present a novel image segmentation method that produces a set of perceptually meaningful regions. The method is based on a hill-climbing approach and achieves the segmentation by performing two main tasks. First, the hill-climbing algorithm detects local maxima of clusters in the global three-dimensional color histogram of an image. Then, the algorithm associates the pixels of an image with the detected local maxima; as a result, several visually coherent segments (small regions) are generated. The segmentation algorithm is simple, fast and nonparametric. The whole segmentation process is performed without any hand-tuning of parameters. Furthermore, this method does not assume any a priori knowledge on the number of clusters or the content of an image. Keyword Image Segmentation,Hill-Climbing