Image Segmentation And Target Tracking Based On Improved Mean Shift Algorithm

In the field of computer vision,image segmentation and moving target tracking have important research significance in intelligent monitoring,robot navigation,intelligent transportation,video content analysis and understanding.This paper mainly focuses on the image segmentation and target tracking based on Mean Shift algorithm.In the aspect of image segmentation,a Mean Shift image segmentation algorithm based on quadratic kernel function is proposed,and the influence of different kernel function on image segmentation result is analyzed.In the aspect of target tracking,an improved adaptively Mean Shift target tracking algorithm is proposed by introducing local Binary Pattern(LBP)and weight moment image.It obtained more accurate scale direction estimation.At the same time,the new target model is constructed by combining the Speed-up robust features(SURF)with the target color feature,which further improves the robustness of the algorithm.The main work includes the following aspects:(1)Concerning the problem that over-division and under-segmentation in some image segmentation algorithms based on mean shift,a quadratic kernel-based Mean Shift image segmentation algorithm was proposed and tested on three standard images in this paper.The algorithm firstly uses the kernel function to weight the sample point;then,the gray value of each pixel is iterated to the maximum gray probability density by using the Mean Shift vector;next,the iterated gray value is assigned to the current pixel;finally,the image is segmented by traversing and clustering the current pixel.In addition,the proposed algorithm was compared with the Mean Shift segmentation algorithms based on other four kernels.And the proposed algorithm is applied to the more complex medical image segmentation field.The experimental results show that the proposed algorithm can achieve better image segmentation results.(2)A robust adaptive tracking algorithm for scale direction is proposed for the problem of target scale change under different tracking scenes.The algorithm introduces LBP feature and weight moment image into the framework of mean shift algorithm.The target model is described by combining the LBP feature and the color histogram,which enhances the identification ability of the target model.In addition,we propose a correction function associated with LBP feature and Bhattacharyya coefficient to adaptively adjust the target region.We evaluate the performance of thisalgorithm via application to a variety of test cases and comparison with the widely used EM-shift algorithm and SOAMST algorithm.The experimental results verify that the proposed algorithm can obtain more accurate target location and scale estimation while guaranteeing the computational efficiency.(3)Concerning the problem of sensitivity to illumination change,a new target model representation method is proposed.The SURF feature in target region was extract for its insensitive to illumination change to combine with color feature,and then form a new joint histogram to describe the target model.In addition,the proposed algorithm is applied to the target tracking in scene with strong light changes,and compared with some other classical algorithms.The experimental results verify that the algorithm proposed in this paper enhances the robustness to the illumination changes.