Research On Algorithms For Image Segmentation Based On Fuzzy Theory

Image segmentation serves as a key in image analysis and pattern recognition and is fundamental step toward low-level vision, which is significant for object recognition and tracking, image retrieval, face detection, and other computer-vision-related applications. It is just to divide an image into different sub-images with different characters and extract some interested objects. Recently, many researchers have proposed various segmentation methods based on the new mathematic theories, such as threshold technique, the edge detection, the region growth, neural network and spatial clustering and so on. The fuzzy c-means (FCM) algorithm is one of the most widely used method for data clustering, and has been successfully applied to feature analysis and classifier designs in fields such as astronomy, geology, face detection, medical imaging and target recognition and tracking.Firstly, a novel fuzzy C-means algorithm based on two-dimensional similitude degree is proposed. Traditional fuzzy c-means (FCM) algorithm requires some prior knowledge to determine initial parameters, but the selection of initial parameters usually influences the clustering result greatly. In this paper, more accurate original cluster centers were acquired before fuzzy clustering, which could avoid poor clustering results caused by wrong cluster centers. Moreover, traditional clustering algorithms which do not fully utilize the spatial information are sensitive to the inhomogeneity of color and noise. The distance of the objective function in this paper is defined as the product of the feature distance and the adaptive spatial distance with different weight. The similitude degree of feature information and spatial information between pixel and cluster center were used in the process of fuzzy clustering. A novel objective function has been established which contains neighbor information and punishment function. The experimental results show that the method has stronger anti-noise property and higher segmentation accuracy.Secondly, an adaptive fuzzy C-means algorithm for image segmentation based on index of fuzziness is proposed. It automatically determines the proper number of fuzzy clustering by utilizing the gradient detection method of wave trough and peak. Accurate original cluster centers were acquired by utilizing fuzzy threshold method. A novel objective function has been established which contains feature information and spatial information. The experimental results show that the method has strong anti-noise property and high segmentation accuracy, and the speed of it is fast.