Abstract. We introduce a new approach to image segmentation, which is based on modeling the spatial distribution of points in the image plane and their ability to identify clustering. Based on histogram peaks detection, a sequence of dominant intensities is formed for each fragment. For each image intensity, point patterns are formed. Then, their ability to cluster the point patterns, formed for each images intensity, is checked, based on the second-order characteristics of the point pattern. The union of all intensities, which corresponds to the clustering point patterns, forms the object of segmentation. The results of image segmentation in comparison with thresholding and seed region growing methods are shown.

Keywords: image segmentation, clusterization, point field, spatial distribution, local extremum.