Research On Robust Segmentation Algorithm Based On Semi-Supervised Fuzzy Clustering

Image segmentation is an important issue for image analysis and pattern recognition,the segmentation results directly affect the subsequent quality analysis of the image and pattern discrimination results analysis.The essence of image segmentation is the clustering of pixels,pixels are classified based on the similarity between the pixels,so that the similarity of pixels in the same class is as high as possible,and the similarity of pixels in different classes is as low as possible.Fuzzy clustering algorithm is a classical unsupervised clustering algorithm,which can reasonably describe the fuzziness and uncertainty of pixel ownership,so it is widely used in image segmentation.Among them,fuzzy c-means clustering(FCM),as a typical algorithm,has been studied and improved by many scholars due to its advantages of easy implementation and fast convergence speed.Semi-supervised learning is a research hot spot in the field of machine learning and data mining.With image segmentation as the application background,semi-supervised learning can fully exploit and utilize the prior information of the image to make the segmentation result more efficient and accurate.Therefore,many scholars are attracted to study the field and achieved certain research results.Among them,the semi-supervised FCM algorithm based on the objective function embeds the semi-supervised regular terms of prior information in the objective function of the FCM,so that the clustering process can proceed smoothly under the guidance of prior information.In this dissertation,the existing problems of the FCM algorithm and the semi-supervised FCM algorithm are systematically studied,the main research contents are as follows:As color image carries more visual information,color image segmentation has become an important research area.Facing the application of color image segmentation,the fuzzy clustering algorithm based on Mahalanobis distance is used.On the basis of this,the noise pixels are reconstructed by using the information of neighboring pixels to improve the anti-noise performance,and the 3D-histogram is constructed by means of three channel components of color image in RGB space to fast segment color image,then a fast robust clustering algorithm for color image segmentation is proposed.For the situation that the image to be segmentation is disturbed by noise or pixel data is insufficient,the average of priori information of the pixel category is used to guide the clustering process to a good search space,and the means of neighborhood image pixels is embedded in the objective function of semi-supervised fuzzy C-means clustering algorithm,then a semi-supervised clustering iterative expression with spatial information constraint is obtained by the optimization method.At the same time,the local Gauss kernel function is used to map the pixel samples from the Euclidean space to the high dimensional kernel space,and then the kernel space semi-supervised fuzzy clustering segmentation algorithm with spatial constraints is proposed.In order to further improve the performance of gray image segmentation and noise resistance,two improved semi-supervised FCM cluster segmentation algorithms are proposed.Firstly,combining the spatial information,gray information and non-local geometric structure information of the pixel neighborhood,a weighted distance is defined to replace the distance measure in the semi-supervised FCM objective function,then a high-performance robust semi-supervised cluster segmentation algorithm is obtained using the optimization mathematical method,and its convergence is proved.Secondly,a semi-supervised fuzzy clustering segmentation algorithm combined with guided filtering is proposed.On the one hand,the spatial neighborhood weighted distance is used to measure the dissimilarity.On the other hand,guided filtering is used to preprocess the noisy image,and the result of filtered pixels is used as a prior information of pixels,which is combined with the prior information of pixel membership degree,so as to constraint the semi-supervised clustering objective function.The improved semi-supervised FCM algorithms can effectively segment gray images under strong noise interference,which can meet the requirement of image segmentation.