LBP and Its Variations for Image Classification
Lei Guo, Qingshan Li · International Conference on Electronics, Communications and Control · 2012
The local binary pattern (LBP) is a simple effective local texture feature and widely used in the fields of face analysis, motion analysis, image segmentation, and image & video retrieval et al. However, there is less considering of color information and descriptor shapes in traditional LBP. In this paper, we investigate the performance of LBP feature in six color spaces which are invariant to illumination condition changes. In additional, we extent the LBP to larger area with more pixels and descriptor shapes involved and expect to improve its performance while keeping the computational complexity. Experimental results on the well-known PASCAL VOC 2011 dataset give special preference to the C-LBP, Opponent-LBP and the diamond descriptor with 8 neighbor pixels around center.