Content-Based Image Retrieval Based on Texture Spectrum Descriptors

To improve the performance and decrease the dimension of the texture spectrum descriptors,a novel descriptor,named ICS-LBP,is proposed based on LBP(local binary pattern)and CS-LBP(center-symmetric local binary pattern).Compared to the gray-value difference between the center-symmetric pixels used in CS-LBP and the gray-value difference between the center pixel and its neighbors used in LBP,the new descriptor classifies local patterns with respect to the gray value difference of the central pixel and the center-symmetric pixels,and thus fully uses the texture information of the local area.In addition,the novel descriptor can be refined by comparing the average gray of local regions and the entire image.The descriptors have been tested on three texture databases with different evaluation criterions and the experimental results demonstrate plausible performance for texture image retrieval.