Study Of Image Segmentation By Thresholding Incorporating Spatial Information

Image segmentation is the process of grouping image’s pixels into several meaningful and non-overlapped homogenous regions. Within a region, image’s characteristics, such as gray-level, texture, color are similar while image’s characteristics are different between regions. Thresholding method becomes the most commonly used method in image segmentation, because of its simpleness and utility. But it only uses the gray-level information, which cant’t get ideal segmentation results. The fact shows that using both the gray-level and spatial correlation between pixels help to improve the segmentation results. To improve the accuracy of segmentation result, we fulfill the following work:First, we analysis the shortcomings of the traditional two-dimensional histogram and propose a new two-dimensional histogram, called gray level-local variance(GLLV) histogram. GLLV histogram is composed of the local variance of pixels and gray-level, which depicts the dispersion of grey level distribution in a neighborhood and contains the edge information. Then, optimal threshold values are selected by maximizing the Shannon entropy of objects and background, which are calculated according to GLLV histogram. Experimental results show that entropic thresholding method based on GLLV histogram can get better segmentation performance.In addition, we propose a novel method for constructing the 2D histogram by usinggrey-level and local entropy, called GLLE histogram. It represents the variance of local region and catches the natural properties of transition regions. Then we point out that Tsallis entropy is nonextensive, and it can describe more complex statistical characteristics. Finally a Tsallis entropy thresholding method based on GLLE histogram is presented. Experimental results show that it can improve the segmentation accuracy.We use gray-level and its local fuzzy entropy in a neighborhood to constitute a new two-dimensional histogram(GLFE histogram), by considering that fuzzy entropy can reflect the information related to edges. Based on the GLFE histogram, we can get the ideal threshold to segment the images by using Tsallis’ s entropy thresholding method. Furthermore, a nonlinear fuzzy membership function is used to map the original image to fuzzy domain in the procedure of calculating fuzzy entropy, making it not only suppress noise but also enhance the weak or fuzzy edge. So Integrating fuzzy entropy information into thresholding process can enhance the segmentation performance of images with the weak or fuzzy edges.