ANew Parallel Thinning Algorithm ' For Gray Scale Images

Abstract

A new parallel thinning algorithm for gray scale images is proposed. The algorithm is based on repeated application of removal operation which erodes gray scale image until only a one pixel thick subset is obtained. The removing operation is applied to all pixels in the image in parallel. A small neighborhood binarization is considered for evety pixel. The set of pixels which satisfy certain conditions is removed. The conditions guarantee that the resulting thinned version is connected. Two types of removal operations are considered . Restricted removal and unrestricted removal to guarantee that the resulting thinned version is located along the center of the ridges where ihey exist. The algorithm can process simple gray scale images (images do not contain hollows surrounded by ridge lines) as well as images that contain hollows. A hollow at elevation k is a flat region such that a path from any pixel in the region to a pixel not in the region must include at least one pixel having a value greater than k. Thinning images which contain hollows are accomplished by using a hollow detection procedure which converts an interior pixel into a border pixel so as to render it a candidiate for removal. The algorithm was tested using three groups of images, 40 chromosomes images, 24 actinomyces images and 6 text and graph like Images. Different tuning parameters were used in the tests. These parameters are the number of smoothing iterations, the percentage of edge pixels used in the hollow detection procedure, the type of removal operation which may be restricted or unrestricted and whether the image is segmented (into background with zero value and objects with positive value) or not. The results showed that with the proper selection of the tuning parameters the algorithm is a powerful tool that could be used for image analysis applications