In order to describe the image texture feature efficiently,a new extension of local binary pattern (LBP) is proposed f or texture image retrieval in this paper.Firstly,a new definition of image local edge is introduced according to the gray-level variation between the central p ixel and its neighbors in an image neighborhood.Then,the new method,called local edge binary pattern (LEBP),is presented,which fuses the advantages of LBP and l ocal edge information.After that,the two region operators,center-symmetric loca l binary pattern (CS-LBP) and direction local binary pattern (D-LBP),are impro ved to center-symmetric local edge binary pattern (CS-LEBP) and direction loca l edge binary pattern (D-LEBP) respectively without increasing the computationa l complexity or feature dimension.Finally,LEBP is further extened by combining w ith Gabor filter.In order to prove the performance of the descriptors mentioned in the paper,three widely used texture databases are chosen for test.The experim ental results prove that CS-LBP and D-LBP can be greatly improved if they comb ine with the proposed LEBP.Furthermore,the multiresolution operators can get bet ter performance than the traditional LBP with less feature dimension.