Content-Based Image Retrieval Based on Improved Rotation Invariant LBP Descriptor
Shuai Wang, Yingying Zhang, Mingyu Nie, Yupu Zhao, Zijiang Yang, Shiwei Zhu, Yanqing Zhao · 2019
This paper proposes an improved rotation invariant LBP descriptor (improved rotation invariant LBP, ILBPri) by adding sampling to improve the performance of the classic LBP descriptor description and retrieval efficiency employed in the CBIR (content-based image retrieval) system. Firstly, Harris corners are extracted from the original image by ILBPri descriptor, then the original image is sampled centered on the Harris corners; secondly, the sampled image are encoded in rotation invariant LBP; thirdly, the LBP histograms of each image are counted; finally, images are ranked by Euclidean distance between LBP histograms of images. A CBIR system was designed in this paper to test the performance of ILBPri descriptor, and an image retrieval experiment was designed with test image database Caltech 256. Experimental results show that precision has been increased by at least 1.8% and recall has been increased by at least 2.4% compared with classic LBP and classic improved LBP descriptors. The query time has been increased by 23.78ms and 0.77ms compared with LBP and LBPri.