Application And Research Of Clustering Algorithm In Image Segmentation

Image segmentation is one of the most challenging research topics in computer vision and pattern recognition.It is an important prerequisite for image analysis,image understanding and image description.With the deepening of research,scholars have proposed a large number of image segmentation algorithms,of which the most classic is based on fuzzy clustering image segmentation algorithm,have been widely used in a number of research areas.Because the fuzzy C-means clustering algorithm only considers the gray information and ignores the spatial information of the image,it is more sensitive to the noise.In view of the above problems,a variety of improvement program rest on space information was proposed.This dissertation is mainly focus on the improvement of FCM and its clustering effect.Firstly,the standard FCM algorithm is analyzed deeply,the advantage and disadvantages of fuzzy clustering algorithm are expounded.The improved FCM algorithm was introduced in recent years,and they were divided into two categories,the first category: fuzzy clustering algorithm using local spatial information of image;The second category: fuzzy clustering algorithm using non-local spatial information of image;and the related algorithms are briefly analyzed and verified experimentally.Secondly,according to the problem of slow convergence rate of FCM algorithm,a fast fuzzy clustering algorithm based on combinatorial membership degree was proposed,it divided the membership degree matrix in the iteration by combining the membership function,which avoids the shortcomings of time consumption of calculating the neighborhood information.The membership function not only takes the local information of membership degree into account,but also takes the spatial information of membership degree into account.It effectively reduces the time complexity of the algorithm meanwhile ensures the accuracy of image segmentation.Finally,a fuzzy clustering algorithm based on combinatorial distance was proposed in view of FCM algorithm to ignore neighborhood information.It used the redundant information of the image to calculate the correlation between the pixels in advance.By defining the new combination distance and spatial distance of the pixel to the clustering center,it not only improves the anti-noise of the algorithm,but also enhances the clustering effect of the algorithm.