The Researchand Application Of The Image Segmentation Algorithm Based On Fuzzy Theory

Image segmentation is the basis of image analysis, recognition and understanding. It isthe most important problem to realize automatic image analysis and pattern recognition, andis also one of classical problems of image processing. As color image provides more abundantinformation than gray image, color image segmentation is drawing more and more attentionson itself. Many researchers had proposed all kinds of image segmentation methods. In thispaper, a color image segmentation algorithm based on crisp segmentation, multi-directionfuzzy morphology edge detection algorithm, as well as pixel coverage segmentation based onlinear unmixing and minimization of perimeter and boundary thickness is proposed.Experiments prove that the algorithm outperforms the result of traditional crisp segmentationmethod and FCM clustering method, it can effectively eliminate the noise, and the algorithmcan effectively reduce execution time. By the improvement and integration of the previousalgorithm, a color image segmentation method based on multiple color spaces and pixelcoverage segmentation is proposed. Experiments prove that the algorithm outperforms theresult of FCM clustering method in different color space. It can realize image segmentation indifferent color space. Four images are selected from Berkeley image library. The segmenta-tion results in different color space are evaluated by PRI and VOI. By comparison, the methodin this paper is better than FCM clustering method. Also, PRI and VOI can help to select theappropriate color space for image segmentation. The main job of this paper is as follows:1. For that only the boundary pixels were mixed pixels, a pixel coverage segmentation methodbased on crisp segmentation was proposed. First,segmented the image by a crisp segmentationmethod. Then, extracted the set of mixed pixels by fuzzy morphology edge detection method.In the end, assigned pixel coverage values of mixed pixels by the pixel coverage segmentationmethod. It can reduce the number of pixels to deal with. It can effectively reduce executiontime.2. For that the number of segmentation in the pixel coverage segmentation method needs to bedetermined by human, the number of peaks can be determined by histogram peak method inHSI space. Then, determined the number of image segmentation according to the number ofpeaks and actual situation.3. For the initialization of coverage segmentation matrix in pixel coverage segmentation algo-rithm, using correspongding distance similarity measure in different color space can realizeautomatic initialization of the coverage segmentation matrix.4.The segmentation results of four images which are selected from the Berkeley image libraryin different color space are evaluated by PRI and VOI. The results of evaluation can help toselect the appropriate color space for image segmentation.