Research On Image Segmentation Algorithm Based On Fuzzy Clustering

Image is an important way of human cognition.Image processing is also a very important area in machine learning.Image segmentation technology is a very important part of image processing technology,which is also a research hotspot in current academia.Fuzzy C clustering algorithm(FCM)is a clustering segmentation method widely used in image segmentation.By introducing fuzzy theory into hard clustering,the algorithm can reflect the fuzzy characteristics of the image better and obtain a good segmentation result.However,the FCM algorithm has some problems such as lack of robustness and lack of spatial correlation.In response to these problems,this paper studies the FCM algorithm and proposes an improved algorithm.First of all,this paper analyzes the implementation of standard FCM algorithm.FCM algorithm uses the Euclidean distance as a criterion to measure the similarity between images.Although this method is simple,the Euclidean distance is not accurate in the image segmentation.In response to this problem,we propose a new criterion to replace the Euclidean distance as a new criterion for clustering.The new criterion,based on the HSI color coordinate system,removes the intensity components which have no value to the segmentation results.It combines neighborhood information,and non-linearly converts hue and saturation.At the same time,this paper analyzes the FCM algorithm based on our new criterion,and experiments show that our algorithm has better segmentation results.Afterward,aiming at the lack of space information and the lack of robustness in FCM algorithm,this paper proposes a fuzzy clustering segmentation algorithm,which is based on Markov random field model and combined with spatial neighborhood information.The algorithm introduces the label field prior probability of Markov model into the FCM clustering model,and combines the global spatial texture features in Markov with the spatial neighborhood information.So it can compensate for the FCM model lacking of description of the spatial information.At the same time,the introduction of neighborhood information also reduces the impact of noise points and improves the robustness of the algorithm.By simulating the images of Berkeley dataset,the AP value of this algorithm is higher than other typical improvement algorithms by 2%?4%,so the proposed algorithm is proved to be more robust and more accurate.