Boundary Detection of Textured Regions

Abstract

It is not easy to find exact locations of boundaries of textural regions since statistical properties around the boundary show mixed properties of two neighboring regions. This paper presents a method to detect exact locations of boundaries of textural regions in the scene. We first propose 'a measure of boundary' which shows large value on the boundary and very small value in other locations. Then we propose a method to find globally good boundaries with respect to the defined measure of boundary using the optimization procedure. The initial segmentation is performed on the reduced version of the input image using multiple textural properties by the recursive thresholding method. This segmentation result on the reduced image is then utilized as a plan for detecting exact locations of textural boundaries in full size image by the proposed textural boundary detector. The proposed method has been applied to a number of textured images, and some of the results will be given in the paper.