Effective image segmentation is an important task in computer vision.In view of computational complexity and poor description of the image segmentation by maximal similarity based region merging (MSRM), a novel fast image segmentation algorithm,i.e.,improved MSRM (IMSRM),using local binar y pattern (LBP) to calculate the similarity between the adjacent regions is proposed.LBP texture descriptor,whi ch encodes the local micro-structure between the image pixels to achieve a description of their spat ial relationship,effectively improves the description capability of the region feature,the obtained feature vector dimension is much smaller than the color histogram,and greatly improves calculation efficiency of the adja cent area similarity.The proposed algorithm automatically merges the regions which are over segmented by mean shift algorithm,with the marker indicating the region of the object and background.The region merging process i s adaptive to the image content and it does not need to set the similarity threshold in advance.A large number of experiments compared with MSRM algorithm show that the IMSRM algorithm can effectively extract outlin e of the object from a variety of complex backgrounds with better edge details,and the efficiency of the algorith m can be improved by about 50%.