Local binary pattern-based discriminant graph construction for dimensionality reduction with application to face recognition

Abstract

Graph construction has attracted increasing interest in recent years due to its key role in many dimensionality reduction (DR) algorithms. On the other hand, our previous study shows that the Local-Binary-Pattern Image (LBPI) representation is a more powerful discriminant and is invariant to monotonic gray level changes. Here, we attempt to construct a discriminant graph for DR in the LBPI representation space. We call the graph the Local-Binary-Image Discriminant (LBID) graph and further incorporate the LBID graph into the Locality Preserving Projection (LPP) to develop an enhanced algorithm - Local Binary Image Discriminant Preserving Projection (LBIDPP). Meanwhile, we also construct a Local-Binary-Histogram (LBH) graph in LBP histogram space and obtain the Local Binary Histogram Preserving Projection (LBHPP) algorithm and compare these to the LBID graph and LBIDPP. It is worth noting that LBIDPP is not a simple combination of the two feature extractions LBP and LPP, i.e., LBP + LPP. LBIDPP inherits the attractive properties of the LBP and LPP. The experiments on face recognition validate the effectiveness and feasibility of the LBID graph and LBIDPP.