Research For Image Segmentation Based On Fuzzy Theory Analysis

Image segmentation is a basic technology of computer vision, which is the key stepfrom image processing to image analysis. During imaging, due to the influences of allkinds of factors, the image exist many uncertain and inaccurate factors. However, thefuzzy theory can well describe these uncertainties of the image. In recent years, imagesegmentation based on fuzzy theory has been widely studied. This thesis mainlyoverviewed the classical fuzzy clustering algorithm and the fuzzy active contour, andanalyzes the existing problems of image segmentation algorithm based on fuzzy thoery.And an in-depth study on this problem is done in this thesis. The new effective methodshave been designed for some actual problems of the fuzzy clustering applied to imagesegmentation.The main works in this thesis as follow:1. For the problems of the bad segmentation region consistency and the segmentationof the low accuracy in the image containing high strength noise, we proposes a newfuzzy c-means clustering algorithm of image segmentation based on neighborhoodspatial information in this paper. The proposed algorithm introduced the trade-offweighted fuzzy factor, considering pixels space distance constraints and spatial graylevel constraints. Simultaneously, it guaranteed the integrity of the neighborhoodinformation. Next, we introduce a kernel distance measure to its objective function,which can accurately reflect the distribution characteristics of data, improving thesegmentation accuracy. Furthermore, the proposed algorithm avoids the problem of theparameters selection and promoting stability and clustering performance of theclustering algorithm. Experimental results on synthetic and real images show that thenew algorithm is effective and efficient, and is relatively independent of the type ofnoise.2. This paper also puts forward an improved fuzzy active contour model based onneighborhood information for uneven gray image segmentation. The existing fuzzyactive contour model only considered the global information of gray level image. So theimproved fuzzy active contour model make full use of gray level uneven image localstatistical information, by introducing the gaussian kernel function to control theeffective scope of neighborhood window. Experiments have shown that the improvedfuzzy active contour model for uneven gray image can get clear and accurate segmentation results and improve the accuracy and efficiency of image segmentation.