A Combined GLQP and DBN-DRF for Face Recognition in Unconstrained Environments

This paper proposes a novel approach for accurate and robust face recognition by using Local Quantized Patterns computed from gabor-filtered images(GLQP) and Deep Belief Network ensembled dynamic random forests(DBN-DRF). GLQP is a kind of local pattern feature extractor based on gabor filters applying, it makes use of vector quantization and lookup table to let local features become more expressive without sacrificing simplicity and computational efficiency. DBN-DRF is a new deep architecture we proposed, in which dynamic random forests classifier is employed to replace inherent Softmax classifier or SVM to achieve a decent classification result at the top of network. GLQP exploits low-level local features that are used as input to DBN-DRF, which further extracts high-level abstract features for classification. Our architecture is trained and evaluated on two challenging face recognition datasets(FERET and LFW), the experiments result show our approach is competitive or better than the state of the arts.