Image Segmentation Based On IB Method

Image segmentation is an important step during the procedure from the image processing to the image analyzing. It has always been the concern of the majority of researchers. The quality of image segmentation directly influences the results of image analysis. The traditional image segmentation method always requires users have a wealth of prior knowledge and it often falls into the local optimal solutions. In this paper, we introduce the IB method into image segmentation and pick up a new method called IS IB to solve the above problems effectively.This paper presents an image segmentation algorithm, called ISIB, based on the Information Bottleneck (IB) method. ISIB extracts the image patterns by maximally preserving the mutual information between the segments and the gray scale values. There are two stages in our algorithm, partitioning the image and merging the segmentations. In the partition process, we segment an image by maximizing the mutual information gain, so that the fine structure of the image can be obtained. In the second stage, we use the density based IB method to merge the fine segments to get the whole structure of the image. Our experiments show that, compared with other advanced image segment methods, ISIB induces the contours which better describe image objects.