Application Of Mean Shift Algorithm In Color Image Filtering And Segmentation

Images are often polluted by various noises in image acquisition and transmission, which lead image qualities to degradation in some different extent. To improve the image quality, noise is required to be reduced or suppressed. Additionally, image segmentation is a basic step to image analysis and image understanding to find useful features. Compared with gray images, color images may provide human visual system with rich color information and more details. Hence, it is important to research some new filter and segmentation methods on color images in the view of theoretical value and practical significance.Mean Shift algorithm is a non-parameter estimation method based on kernel density gradient. As an effective statistical iterative algorithm, many researchers have paid more attention to it since it is suggested, and currently it has been wildly-used in image processing and computer vision. This dissertation focuses on the application of Mean Shift algorithm in color image filtering and segmentation, whose main work can be summarized as the following:(1) A new filtering method based on adaptive Mean Shift algorithm is proposed by analyzing objectively two key parameters——bandwidth and weights in Mean Shift algorithm. First, an adaptive spatial bandwidth is decided in terms of gray relational degree of color information. Then an adaptive range bandwidth is calculated based on color coarseness. Additionally, the weight of sampling points is computed in the light of color difference. Finally, the filtered value of current point is gotten by the above adaptive Mean Shift algorithm. Experimental results show that the proposed method can not only effectively suppress noise, provide smoothing effects consistent with the human visual perception, but also preserve more details information.(2) To solve the over-segmentation on color image in traditional Mean Shift algorithm, a segmentation method is proposed which is based on Mean Shift algorithm, region growing algorithm and the definition of superpixel. Firstly, Mean Shift algorithm is employed to initially segment the color image and eliminate the very small superpixel regions in order to obtain some superpixels. Secondly, the superpixel is regarded as a seed region to grow from left to right, and up to bottom according to some similarity criterion. Finally, some small segmented regions are merged. Results of our experiments show that the proposed method can segment color images effectively.