Cross Local Gabor Binary Pattern Descriptor with Probabilistic Linear Discriminant Analysis for Pose-Invariant Face Recognition

Automatic face recognition is a well researched area, but still many of the current face recognition methods sensitive to lighting and pose changes. In this paper we introduce a novel facial feature representation to enhance the robustness of face recognition against pose and illumination changes. Here we combined Gabor wavelets and Cross Local Binary Patterns for facial feature representation. Gabor wavelets are known to extract shape information by detecting shape attributes like edges, corners and blobs. Cross Local Binary Patterns can be used for better feature representation in two levels. Probabilistic Linear Discriminant Analysis (PLDA) minimizes the intra-class distance and maximizes the inter-class distances and generates a model for classification purposes. During recognition, PLDA estimates the likelihood of the probe image in the gallery image set. Experimental results on YALE, FERET and our internal datasets show the significance of this method.