Pairwise Cross Pattern: A Color-LBP Descriptor for Content-Based Image Retrieval

The local binary pattern (LBP) has been widely considered an excellent and extensive feature descriptor, but it is limited to grayscale image processing. Inspired by human visual system, we develop a novel yet simple rotation-invariant color-LBP descriptor-pairwise cross pattern (PCP) to extend LBP to color image processing. In the proposed descriptor, the color information map is firstly extracted using a multilevel color quantizer which is designed based on a color distribution prior in the L* a* b* color space. Then, the color information and LBP maps are paired in parallel to construct a pairwise cross pattern, which is easily extended to the uniform pairwise cross pattern (UPCP) and the rotation-invariant pairwise cross pattern (RIPCP). Finally, compared to numerous state-of-the-art schemes and convolutional neural network (CNN)-based models, the experimental results illustrate that the proposed method is efficient, effective and robust in content-based image retrieval task.