Authors: Weiguo He
Addresses: School of Informatics, Guangdong University of Finance and Economics, Guangzhou – 510320, China
Abstract: The local binary pattern (LBP) is an effective facial descriptor which has been successful as it applies to face recognition. This paper presents an improved LBP with a pyramid model and in face recognition applications, in which a separate output label for each uniform pattern and all non-uniform patterns is reclassified instead of collecting them into a single bin. Firstly, each pixel in the low spatial pyramid is obtained by down sampling from its adjacent low-pass filtered high-resolution image. Secondly, for each pyramid, the face area is divided into small regions from which those pixels whose patterns is 'non-uniform' are further processed to extract their LBP pattern with a smaller radius, and replace the non-uniform pattern with the pattern in a small radius. In the final radius tier, those non-uniform patterns are incorporated into existing uniform patterns by minimising the Hamming distance between this pattern and the uniform patterns. Finally, all histograms are concentrated into a single, spatially enhanced feature vector to be used as a face descriptor, in which all patterns are uniform. The experiment on the ORL face dataset, as well as the Honda/UCSD video database shows that the proposed scheme is superior to other related methods.
Keywords: local binary pattern; LBP; uniform patterns; feature extraction; face recognition; computer vision; pyramid model; Hamming distance.
International Journal of Communication Networks and Distributed Systems, 2014 Vol.13 No.3/4, pp.380 - 390
Received: 02 Sep 2013
Accepted: 18 Apr 2014
Published online: 30 Aug 2014 *