Image retrieval based on an improved CS-LBP descriptor

The texture spectrum descriptors, center-symmetric local binary pattern (CS-LBP) and its extension D-LBP are effective for region description. However, they discard the low frequency information of a region. A novel improved operator, named ID-LBP, is proposed based on D-LBP in this paper. The new operator classifies the local pattern based on the relativity between the local gray mean and the center-symmetric pixels instead of the gray variation between the center-symmetric pixels and the center pixel as D-LBP. Comparisons are given among CS-LBP, D-LBP and ID-LBP on two commonly used image databases and the experimental results show the performance improvement of the new method.