Research On Image Segmentation Method Based On Fuzzy Clustering

Image segmentation is a process,which is to divide a image into some areas continuous not overlap with similar characteristics(such as gray scale,color and texture information,etc.)and extract the specific targets from complex background.Because of image's uncertain and fuzzy characteristics,traditional image segmentation methods usually can't achieve ideal results,which affects the later image analysis.Fuzzy theory can express the image characteristics,so it is applied to image segmentation by a lot of scholars,and many new segmentation methods are proposed.Among of those,fuzzy c-means clustering(FCM)algorithm is the most popular,but FCM has some drawbacks,for example,it is sensitive to initial cluster center and easily traps into local optimum,and is also sensitive to noise because of the pixel's spatial neighboring information not taken into account.In the thesis,relative modified methods are proposed to address the problems aforementioned by analysis and improve the performance.The main research work in the dissertation is as follows:(1)To solve the problem about how to select the initial cluster centers,a fast fuzzy c-means clustering(FFCM)method is proposed based on artificial bee colony(ABC)for image segmentation.In the method,ABC which has strong global search ability is used to optimize the initial cluster center of FFCM.However,ABC has some drawbacks,such as being easily trapped in local optimum and slowing convergence speed in the late.So Boltzmann selection strategy is introduced to replace roulette wheel rule,and opposition-based learning strategy is applied to generate new nectar and update the worst nectar in ABC,which can avoid premature convergence and improve the convergence speed.Finally,the modified ABC is applied to optimize the initial cluster center of FFCM,and which overcomes the problem of sensitivity to initialization.Segmentation experiments show that the proposed method can not only achieve the optimal cluster center by less iteration,but also has bettet effectiveness of image segmentation and better performance.(2)A kernel-based fuzzy c-means clustering method is proposed with spatial neighboring information for image segmentation to deal with the problem of the traditional FCM which is sensitive to noise because of spatial neighboring information not taken into account.The method adds a spatial constraint item and defines a spatial neighboring membership function with neighboring information under consideration by priori probability of each pixel,and it replaces the original Euclidean distance with kernel-induced distance to optimize the character of the image sample.Finally,a new membership function with trade-off is generated by incorporating global and spatial neighboring membership functions,and which expands the weight of neighboring information in the clustering.The experimental results show that theproposed algorithm has superior performance in term of quality and effect on image segmentation than traditional FCM,KFCM(Kernel-Based Fuzzy C-Means)and some other modified FCM-based algorithm,and performs more robust to noise.