Based on the analysis of the local binary pattern (LBP) and its extensions,a novel method,called concave-convex partition (CCP),is proposed in this paper to improve the performance of the LBP -based methods for rotation invariant texture classification.By the CCP,the neighborhoods of the image are divided into two categories firstly,the concave and convex categories, before computing the local binary codes.The reason is that the neighborhoods wi th different structures and visual perceptions may be set the same LBP code by t he traditional LBP-based methods,which can reduce their discriminability inevit ably.Then,two histograms are built on the concave and convex categories,respectively and concentrated into on e as the texture image feature.Experimental results obtained from three widely used texture image databases demonstrate that the pro posed method can greatly improve the performance of the traditional LBP-methods on texture classification.