Search

In the extraction process of LBP features, most consumption of time and memory were paid for clustering. In order to address these problems, an improved local binary algorithm for image categorization was proposed. The algorithm replaced decimal system encoding LBP with binary descriptor. Meanwhile, Hamming distance was employed rather than Euclidean metric for features clustering. The multi-scale LBP features was flued for a new local binary descriptor. The result of the experiment on the PASCAL VOC 2007 dataset showed that the adopted local binary descriptor was better than the classical LBP, specifically for time consumption. [ABSTRACT FROM AUTHOR]