Novel similarity measures for face representation based on local binary pattern

ZHU Shi-hu, FENG Ju-fu



Author NameAffiliation

ZHU Shi-hu Key Laboratory of Machine Perception Peking University, MOE Department of Machine Intelligence, School of Electronics Engineering and Computer Science, Peking University, Beijing 100871, China, zhush@cis.pku.edu.cn, fjf@cis.pku.edu.cn 

FENG Ju-fu Key Laboratory of Machine Perception Peking University, MOE Department of Machine Intelligence, School of Electronics Engineering and Computer Science, Peking University, Beijing 100871, China, zhush@cis.pku.edu.cn, fjf@cis.pku.edu.cn 



Abstract:

The successful face recognition based on local binary pattern (LBP) relies on the effective extraction of LBP features and the inferring of similarity between the extracted features. In this paper, we focus on the latter and propose two novel similarity measures for the local matching methods and the holistic matching methods respectively. One is Earth Mover’s Distance with Hamming and Lp ground distance (EMD-HammingLp), which is a cross-bin dissimilarity measure for LBP histograms. The other is IMage Hamming Distance (IMHD), which is a dissimilarity measure for the whole LBP images. Experiments on FERET database show that the proposed two similarity measures outperform the state-of-the-art Chi-square similarity measure for extraction of LBP features.

Key words:  similarity measurement  local binary pattern  Earth Mover’s Distance  IMage Euclidean Distance

DOI：10.11916/j.issn.1005-9113.2009.02.015

Clc Number:TP391

Fund: