A Study Of Segmentation Algorithm For Images With Fuzzy Boundaries

In the field of images research and application, images segmentation has always been a difficult and hot problem. While fuzzy boundary images segmentation, being influenced by the fuzzy boundary, is even more difficult. This thesis aims to find an effective algorithm for fuzzy boundary images segmentation.The thesis first adopts some representative and commonly-used algorithms, including fuzzy C-means clustering algorithm, active model snake contour detection algorithm and canny operator edge detection algorithm, to conduct some applicational segmentation experiments on fuzzy boundary images, in order to test the applicational effects of these algorithms, and also to analyze their merits and demerits. Through the above research and analyses, and on the basis of region growing, this thesis proposes two algorithms concerning fuzzy boundary images segmentation.The first algorithm is the regional integration algorithm based on gradient boundary. The starting point of this algorithm is as follows:the boundary in an image is always caused by the pixel difference between adjacent regions, and the bigger the difference, the higher the credibility of the boundary. This algorithm first uses region growing method to generate the fuzzy boundary image into several regions, regards the whole image as a three-dimensional map with the pixel value as altitude, and then finds the maximum-gradient boundary in the map through iteration; in the meantime, combines the slope principle to smooth the regions on the same slope, and gradually segments the image into different target regions. Experiments indicate that, in comparison with the above commonly-used algorithms, this algorithm can better meet people's demands for image observation.The second algorithm is a fuzzy boundary images segmentation algorithm based on fuzzy connectivity. The principle of this algorithm is based on the process of human eyes percepting objects:if the pixel informations of two regions are similar enough, then human eyes will see them as two parts of one and the same object. This algorithm starts from inspecting the fuzzy connectivity between each connected region, and treats the regions within certain range of fuzzy connectivity as partial regions of a certian object, and then merges the adjacent sub-regions within certain degree of fuzzy connectivity; through this continuous iteration and aggregation, hence gradually and finally segments the objects from the fuzzy images. The segmentation experiments on several images in this thesis indicate that, this algorithm can effectively segment fuzzy boundary images into different target regions.After comparing and analyzing the segmentation effects of these algorithms mentioned in this thesis, it is found that, the two newly-proposed algorithms are obviously superior to other segmentation algorithms in terms of segmentation effect. In particular, the algorithm based on gradient boundary can better determine the authenticity of boundary, while the algorithm based on regional fuzzy connectivity, is able to distinguish the integrity of the objects and to segment the images at many levels by altering the membership degree of fuzzy connectivity, hence to find target regions in the images at various levels.This research bears the following innovative points:1) employing the gradient boundary thought to determine the effective boundary, according to the pixel characteristics of different regions; 2) using slope smoothing method to merge regions with small characteristic difference, thus to eliminate false boundary; 3) first marking the ambiguous fuzzy regions (which may either be the object or the background), then after the segmentation of other regions, classifying these marked regions according to their environmental information; 4) deciding the fuzzy connectivity degree by calculating the degree of regional fuzzy affinity; 5) adopting the fuzzy sub-sets segmentation method on the basis of fuzzy connectivity degree, and gradually iterating to accomplish the fuzzy boundary images segmentation.The last part of the thesis makes a conclusion and proposes some prospects. It sums up the merits and demerits of the fuzzy boundary images segmentation algorithms mentioned in this thesis, and puts forward new exploration directions for further improvement. This thesis hopes for some more effective and more excellent segmentation algorithms in future researches.