Research On Fuzzy Clustering Algorithm Based On Deep Fusion Prior

The core idea of fuzzy clustering algorithm is to put the data points with the same characteristics into the same category according to the fuzzy membership,and put the data points with different characteristics into different categories.Because fuzzy clustering algorithm has the characteristics of simplicity and low sensitivity,it can help people find useful information in the data.It is a very popular and mature algorithm in the field of image segmentation.Although many improved algorithms based on traditional fuzzy clustering have been widely used in image segmentation.But the traditional fuzzy clustering algorithm still needs some improvement.Firstly,the image segmentation result of traditional fuzzy clustering algorithm is easily affected by outliers.Secondly,these algorithms often lead to over-segmentation because of the loss of image local spatial information.Therefore,it is difficult to solve the existing problems simply relying on the fuzzy C-means clustering algorithm.Researchers have gradually studied many related methods driven by prior knowledge to improve the performance of image segmentation.However,the traditional prior knowledge of manual design is difficult to guarantee the effect of subsequent segmentation of fuzzy C-means clustering algorithm,which is prone to under-segmentation and the segmentation result is susceptible to the influence of outliers.In order to solve the above problems,this thesis proposes a deep superpixel prior fuzzy C-means clustering with weighted local entropy for image segmentation.Firstly,the concept of weight local information entropy is introduced to minimize the discreteness within classes and maximize the weight information entropy between classes,so as to suppress the influence of outliers in images on segmentation results.In order to solve the problem of under-segmentation of existing algorithms based on prior knowledge and the problem of high computational complexity due to local spatial information merging,this thesis constructed a fuzzy C-means clustering framework with deep superpixel prior and a fuzzy C-means clustering model with double data items.At the same time,the deep network superpixel segmentation algorithm and the traditional superpixel segmentation algorithm are introduced for knowledge fusion.It can reduce the overall complexity of image segmentation algorithm and improve the image segmentation effect.In order to test the superiority of a depth prior automatic fuzzy clustering algorithm based on superpixels proposed in this thesis,more than ten groups of detailed comparative experiments are conducted on the images from the Berkeley segmentation Benchmark dataset.To evaluate the performance of different algorithms for image segmentation,the thesis consider the variation of information(VOI),the boundary displacement error(BDE),the probabilistic rand index(PRI)and the global consistency error(GCE),as the performance metrics.By analyzing the experimental results,the superiority of the depth prior automatic fuzzy clustering algorithm proposed in this thesis can be directly demonstrated.