Image Semantic Segmentation Based On Textons

Image semantic segmentation is to segment an image into a certain semantic meaning of divided blocks corresponding semantic identification for each block segmentation,which is the key technologies of image compression and image retrieval based on contents and other in the field of multimedia applications. The image semantic segmentation research is very active, combined the image segmentation and recognition with the image understanding.And it is also the current domestic and foreign research frontier in the field of image processing. But how to effectively obtain the edge information between different semantic object is still a challenging probl em when there are many kinds of different shapes and sizes of objects in the image.Some of the current image semantic segmentation algorithm s only do the semantic training and prediction by the global features or local pixel sampling of image. Its feature semantic training is inadequate,leading to object contour fuzzy segmentation and causing error identification. A part of the image semantic segmentation algorithms do the semantic training and prediction by the pixel level. Semantic prediction process is very time-consuming while it is possible to ensure enough feature training.For the advantages and disadvantages of these two kinds of algorithms,this paper proposes a kind of image semantic segmentation algorithm based on textons in the full use of the relationship between adjacent pixels and reserving the edge information between objects.This algorithm extracts the texton feature before it uses k-means and k-d tree to get a textons result.And then it can realize semantic mapping to textons by the image semantic training and prediction method based on SVM. Finally it completes the image semantic segmentation.This paper selects 11 sest of objects from MSRC image gallery to experiment. The experimental results show that the algorithm is able to a variety of semantic object segmentation and recognition with a clear outline. Recognition accuracy has a 6.73% improvement compared with the image semantic segmentation algorithm based on high-order CRF model. Prediction efficiency fell to 1.173 seconds per image from 11.657 seconds per image of Texton Boost system based on pixel level,which effectively improve the efficiency of semantic training and prediction.