An Extension of Texture Spectrum Using Local Structure and Variance

In order to improve the performance of the region texture spectrum descriptors, an improved operator, called ID-LBP is discussed firstly. Then, three rotation invariant texture descriptors, named CS-LBPV, D-LBPV and ID-LBPV, are proposed in the paper based on local structure and the variance of local image. The new operators characterize the local contrast information into the one-dimensional texture spectrum histogram and they do not need any quantization. The descriptors mentioned in the paper are tested and compared with the traditional methods on two commonly used texture databases and the experimental results show the performance improvement of the new methods in texture image retrieval.