Research And Application Of The Image Segmentation Method Based On Superpixel Generation

Image segmentation, as one of the most important processing stages of image processing, which according to image texture information, color information, gray value, spatial location information, dividing the image pixel into meaningful foreground object region and background. This is a kind of regional partition problem.In all kinds of pre-processing segmentation method, the super pixel generation algorithm is used as a representative one, with good segmentation performance. According to certain features, the super pixel generation algorithm divides the pixels in the image into different sub region blocks. The early super pixel generation algorithm was mostly aimed at gray image, and sub region of the segmentation is unsystematic, not better play its performance, so that the effect of the segmentation is not ideal. In this paper, we use the two-step segmentation method to segment the image.In order to solve the defects of the traditional super pixel generation algorithm, the improved Watershed super pixel algorithm is applied to the color image directly. For further improve super pixel algorithm performance and maintain positional information of images, we adopted SLIC method, utilized the generated super pixels blocks having equal size and controllable quantity characteristics, aimed at the imprecision problems of the super pixel segmentation boundary, improved the super pixels distribution grid, so that it can fit the objects boundary better. For accomplish extracting the target region, combine region merging strategy. Take the real brain CT image data sets as an example and contrast to other segmentation methods. The experimental results show that the improved super pixel generation algorithm proposed in this paper, its effect has improved significantly.The main research work in this paper is as follows:(1) In this paper, the types of image segmentation methods are introduced, focus on introducing the principle of several traditional algorithms of super pixel generating.(2) Aim at the image information's complex background and uneven illumination problems, improved the Watershed super pixel, proposed the W_MSRM algorithm, and solved the limitation in the traditional Watershed super pixels for color image segmentation.(3) In order to solve the problem of super pixel clustering segmentation inaccuracy, this paper improved the clustering super pixels, and proposed the ISG_MSRM algorithm.(4) The ISG_MSRM algorithm, which we proposed in the previous step (3), is applied to brain CT image segmentation. Then try to delineate and extract the soft tissue temporal lobe region, to provide an effective basis for expert rapid diagnosis.