Image Segmentation Based On Spatial Random Tree

Image segmentation is a key technique in image processing. It is a basic preparatory step in the most problems of image pattern recognition and scene analysis. The quality of image segmentation has effect on the understanding of the image directly. Currently, there are many image segmentation algorithms, but none of them can be generally applicable to variety of images. Therefore, in the field of image processing, image segmentation is still a hot spot.In this paper, we use spatial random tree (SRT) model to split the image recursively and propose an image segmentation method based on SRT and Gabor features, combinating Wedgelet and Gabor features.We use SRT model to split the image, and propose a new split stopping criterion, that is, considering the structure information of blocks in the process of splitting, we stop splitting in case of that the structure of two blocks are similar. Comparing with other traditional partitioning methods, our method can reduce the phenomenon of the over segmentation, maintain regional consistency well and split the edge into a block, the size of which is appropriate.After splitting the image into blocks, we deal with the smooth blocks, texture blocks and edge blocks respectively. First, we merge the adjacent smooth blocks according to gray similarity. Secondly, texture blocks can be get after clustering texture blocks and edge blocks using Gabor features. Finally, we can get the edge location using Wedgelet approximation and merge the two wedge blocks into the appropriate area respectively.Experimental results show we can split image into block reasonably using SRT model. Gabor features can effectively distinguish the texture blocks and edge blocks. Our method can get better segmentation, and is a good compromise in regional consistency property and boundary to property.