Research On Image Segmentation Based On Fuzzy Clustering Algorithm

Image segmentation is one of important research topics in computer vision.And it is a key link in the image processing and image analysis.Therefore,it has high research heat and application significance.As a tool of pattern recognition and machine learning,fuzzy clustering algorithm has been improved and applied to image segmentation in recent because of its simple theory and procedure.However,the traditional fuzzy C-means(FCM)algorithm has many drawbacks such as poor denoising ability,easy loss of details,low efficiency and so on.To improve the robustness and efficiency,this paper proposed an improved fuzzy C-means algorithm based on local information and Markov random field for image segmentation.Besides,we proposed an image segmentation neural network model based on fuzzy clustering label learning for multi-modality brain MRI image.The main work of this paper includes:(1)Propose an improved fuzzy C-means algorithm based on local information and Markov random field.The proposed method takes Markov random field(MRF)to build the relationship between central pixel and its neighborhood.And the method adds the MRF into FCM by introducing prior membership and posteriors membership.(a)An adaptive similarity measure based on local information is proposed.With the local neighbor similarity information and prior membership degree,the novel measure describes the relationship between the pixel and the cluster center more robustly,and the impact of neighbor pixels can be determined adaptively according to the prior membership degree,which improves the segmentation performance under noise.(b)A new prior membership approximation in MRF model with local spatial information and local grey information is proposed.The new prior membership approximation is more robust and efficient by using local spatial information and grey information,while the original function is complicated and ignored the benefits of local information.(2)Propose an image segmentation neural network model based on fuzzy clustering label learning for multi-modality brain MRI image:(a)The neural network model based on fuzzy clustering label learning is trained from local feature of pixels as input data and fuzzy clustering results obtained from input images as target output.With the benefits of the clustering algorithm and the neural network,the proposed model can predict faster and has less data limitation.(b)Combine the prediction results to obtain more accurate segmentation results,according to the Dempster's rule of combination.Dempster-Shafer theory is used to combine evidence from different sources to get final belief.The neural network can quickly obtain lots of prediction segmentation results from multi-modality and multi-method.And Dempster's rule is used to combine these prediction results to make segmentation decision.With the advantages of the different modalities and methods,the image segmentation accuracy is improved a lot.The two methods proposed in this paper are tested in the experimental data set.The experimental results show that the methods achieve satisfactory performances and obtain robust and efficient segmented images.