Research On Image Segmentation Of The Mean Shift Algorithm Based On Color-Texture Feature

Image segmentation is an image processing technology, which plays an important role of computer vision and target recognition. It can segment an image into several meaningful certain regions, and the results of segmentation have been widely used in the field of intelligent transportation, image retrieval, and biomedical field. So the segmentation quality influences on further image processing. The Mean Shift algorithm is a region segmentation method, this segmentation method is extremely similar to the analytical characteristics of the human eye, it has a strong adaptability and robustness. There are still many problems to be solved in the process of the Mean Shift method, such as over-segmentation,under-segmentation, running speed and split-efficiency. For these deficiencies, this paper proposes two improved Mean Shift segmentation algorithms, the improved Mean Shift method based on the color and the texture feature method, and the Mean Shift algorithm based on adaptive width.According to the above, the main work of this paper’s as follows:(1) Aimed at the defect of classic Mean Shift algorithm which is vulnerable to texture feature and the unique texture of image, this paper proposed an improved Mean Shift algorithm combined with color and texture features of the image. The color similarity of the current pixel and neighborhood pixels is computed with Euclidean distance and Gaussian function. After employing the Gabor Wavelet Transformation to extract feature in different directions and scale, combined with color similarity, which are added as weight in Mean Shift process. Experimental results show that the proposed algorithm can segment texture images accurately, especially for the color images. Furthermore, the algorithm can efficiently recognize the over water bridge.(2) Aimed at the time consuming, and computing complexity of the Mean Shift segmentation, an adaptive bandwidth Mean Shift segmentation algorithm is presented. The algorithm uses a histogram of the image to estimate the probability density of the image, then the characteristics and the probability distribution of the around pixels are calculated for computing the bandwidth of each central pixel bandwidth. Experimental results show that the adaptive algorithm is a good solution for the low efficiency of the fixed bandwidth Mean Shift method. The adaptive Mean Shift segmentation algorithm also reports the encouraged result of segmentation.