Research On Fuzzy Clustering Based Methods For Infrared Image Object Segmentation

Target detection and recognition of infrared image is one of the key technologies in the modern military weapon system, and the infrared object detection is mainly depended on the image segmentation technology. Therefore, the accuracy and reliability of the image object extraction and recognition are directly determined by the image segmentation accuracy. An image segmentation method based on fuzzy clustering algorithm can be come down to a nonlinear programming problem with constraints. In this thesis, infrared image object segmentation methods based on fuzzy clustering are deeply studied, and the main work and research results are as follows:Firstly, three typical infrared image object segmentation algorithm based on fuzzy theory is introduced and simulated, including a fuzzy c-means infrared image segmentation algorithm, an adaptive fuzzy c-means infrared image segmentation algorithm based on potential function, and a mean shift infrared image segmentation algorithm based on improved FCM. And the advantages and disadvantages of the three algorithms are analyzed. In addition, two objective evaluation criteria for the quality of the segmentation algorithms are presented to provide an objective basis for analyzing the performance of the improved algorithm.Secondly, an adaptive fuzzy clustering algorithm based on multi-threshold is proposed for infrared image segmentation in this thesis. It aims at problems that the fuzzy clustering algorithm cannot adaptively get the reasonable number of clusters and that the segmentation rate of using the algorithm to segment infrared images with larger difference between the object region and the background region is low. The methodology uses a coarse-fine concept to reduce the computational burden required for the fuzzy clustering and to improve the accuracy of segmentation that a fuzzy clustering cannot reach. The coarse segmentation attempts to segment coarsely using the multi-thresholding technique that can effectively remove the false peak interference, then in order to find a finer segmentation result, the coarse segmentation result is clustered by an improved fuzzy clustering algorithm that introduces an adaptive function to get the most reasonable cluster number and that defines a logarithmic function as a measurement of distance. Experimental results show that the proposed algorithm not only can retain the advantages of multi-threshold method in simple and fast, but also can effectively segment the infrared images with larger gray level differences between the background region and the target region.Finally, a fuzzy clustering based level set method for infrared image segmentation is proposed in this thesis. It aims at problems that the fuzzy clustering algorithm is easy to fall into local minimum and that the segmentation rate of using the algorithm to segment infrared images with smaller difference between the target region and the background region is low. The proposed algorithm combines the advantages of fuzzy clustering algorithm and level set algorithm. Firstly, the infrared image is initially clustered by a fuzzy clustering algorithm optimized by simulated annealing algorithm, and then the clustering results are transformed into the initial contour and the evolution control parameters are embedded with neighbor information, so to segment the image by the improved level set method. Experimental results show that the proposed algorithm can ensure that the results of fuzzy clustering are global optimization, and that full advantage of the gray level, edge and the neighborhood information of the image can be taken by combined with level set algorithm. And this method is adapted to infrared images with smaller gray level differences between the background region and the object region.