Gray Theory Applications In Image Segmentation And Effectiveness Evaluation

Image segmentation is a key step from image processing to image analysis, however, up to now, there is not any fast and effective segmentation methods being suitable to all images. So image segmentation has always been a hot research topic in image engineering for many years.After the development of several years, Grey theory has become a new discipline. It is based on the foundation of gray sequence, supported by the grey relational space, and centered on the grey model system. The theory has been successfully introduced to image processing such as image fusion, edge detection, image matching, and etc.This paper applies Gray theory to image segmentation and its performance evaluation and proposes a fast segmentation algorithm which has good performance of noise resistance in grayscale images. Also, this paper suggests a new evaluation method of image segmentation, which combines the method of Forced Decision and grey relational theory. Main work of this paper is summarized in the following.(1) The existing algorithms on image segmentation and its performance evaluation are discussed, whose suitable applications and advantages are analyzed at the same time. Additionally, Grey theory is further researched especially its application in image segmentation and evaluation.(2) A fast image segmentation algorithm is proposed. In this method, an image is taken as a weighted undirected graph first. Then, after the relationship of grayscales and positions in local regions is discussed via grey relational analysis, a grey weight matrix is established, based on which a grey partition function is constructed. Next, the image is binarized with a grayscale that corresponds to the minimum value of the grey function. Experimental results on visible-light images and SAR images indicate that the proposed method, being superior to some existing methods like Otsu and Ncut, not only can segment the images with obvious difference between targets and backgrounds, but also may suppress image noise effectively.(3) A fast and effective method of performance evaluation on image segmentation is proposed. In this method, as far as the specific purpose, application circumstance and research interest are concerned, evaluation measures are selected first. And then, their relative importances are appointed via Forced Decision. Next, segmentation performance are analyzed and sorted by the weighted model of grey relational analysis. Experimental results show that the method is efficient, fast and stable for providing with an overall performance, which is significant to guide the design or the modification of existing segmentation algorithms.