A New Method Based On The Shape From The Shading Shape From The Texture And Wavelet Texture Segmentation

Image segmentation is the most important research filed in low computer vision; it is the key to process image analysis, image understanding and image description, because the quality of image segmentation affects the following analysis, recognition and description.Image segmentation is always the focus in image technology research, and the interesting/investment is more and more in the field. On the other side, image segmentation apply abroad in the real world, every action which want to abstract, measure the image goal can not do without image segmentation. Although, people have done their best to research in the field, but up to now, we can not find a segmentation method that can apply all type of images. Texture segmentation is one of the important fields in image segmentation, much research focus in the field and obtains many good results.Most texture segmentation method can separate two steps. The first step is the texture feature extraction, this is the base for image segmentation, and the right segmentation is based on looking up a group of rational features which descript the image completely. The second step is the segmentation algorithms; the algorithms should class the features quickly and exactly.This paper focuses three sides in texture segmentation:(1) Feature extraction based Shape from shading (SFS). After analysis the method of shape from shading, the paper presents a feature extraction method based on SFS, the method composite the two SFS method, which improve the stability and speed of processing.(2) Multi-feature extraction. Based on the research in texture segmentation, for example statistical method, structure method and time-space method, this paper present a method which fuses the SFS futures, SFT features and wavelet features, then take is as the segmentation features, and get good results.(3) Multi-level segmentation method based on fuzzy. This paper composites a 2-level texture segmentation method which based on two level clustering and space restriction of neighborhood image pixels, after correct the pixels one by one, we can get an exact segmentation result.The paper analysis the common light estimation method and test check efficiency by test.The method which present in the paper have been test by man-made and real world images, test express that the method is a efficient method that can be used in many type of images.