Image Segmentation Algorithm Base On Fuzzy C-means

Image segmentation is dividing an input image into a number of categories of thesame nature according to certain criteria, to extract the part the people are interested in, itis the basis for image feature extraction and recognition and image analysis to understand.Commonly used methods of image segmentation are thresholding method、edge detectionmethod, region growing method、fuzzy clustering method, these methods have their ownadvantages and disadvantages, suitable for all image segmentation method has not beenproposed.Among the fuzzy clustering image segmentation algorithm, fuzzy C-means imagesegmentation method is the most representative, the principle of the method is simpleand can adaptive iterative to get final segmentation results, but the method also hasobvious defects, on the one hand, the need to pre-know the number of categories forimage segmentation and for each type of artificial cluster center, on the other hand, it doesnot consider the spatial information of the pixels, that without consideration of the impactof neighborhood pixels on the classification results. Typical thresholding methods areOtsu method、 maximum entropy method、 minimum error method and maximumbetween-class method. The Otsu method only considers the distance between the twocategories, without considering the distance between the classes, although the newthreshold segmentation emerge in endlessly, but the choice of thresholds is always anunsolved problem. Focus on fuzzy C-means image segmentation algorithm consideringspatial information, to improve the fuzzy C-means image segmentation algorithm and theOtsu method, mainly as follows:(1) propose a new fuzzy C-means image segmentation algorithm used MRF, MRF isa good description of the relationship between pixel and its neighboring pixels point. Thedefects that does not consider space-pixel could been solved with the new method, at thesame time added the weight factor α the objective function, to adjust the impact weightof local neighborhood pixels in segmentation results.(2) Another method considering the pixel spatial information is to join the penaltyterm in FCM target function, gray values of the pixels and its neighbor pixel constitute acertain of sequences, calculate the gray correlation degree of each sequence, set theweight value of the penalty term by the correlation degree, the experiment proved thatthis method is feasible.(3) Maximum Scatter Difference thresholding method is an improved algorithm of Otsu method, this method not only takes into account the between-class variance, alsoconsiders the between-class variance, but the choice of the scatter difference criterionparameter C has not been resolved, which need to be seted manually. introduce the fuzzymaximum scatter difference criteria into the image segmentation, drawing on fuzzyC-means image segmentation algorithm, propose an effective method for solving theparameter C of maximum scatter difference criteria, and achieved good segmentationresults.