Fusion of Multi-directional Rotation Invariant Uniform LBP Features for Face Recognition

Abstract

The LBP (Local Binary Pattern) feature is one of the dominant methods in face recognition. The growth of sampling density of general Uniform LBP operator will improve the precision with the cost of very high-dimensional feature in face recognition. The normally adopted Riu-LBP(Rotation Invariant Uniform LBP) feature is of very low dimension but deteriorate precision due to the lose of direction information. In this paper, we propose feature-level fusion of multi-directional Riu-LBP features. With such simple scheme, the feature dimension is drastically decreased while the precision is comparable or even better than the general Uniform LBP features.