Learning Compact Binary Face Descriptor for Face Recognition - 2015
Binary feature descriptors like native binary patterns (LBP) and its variations have been widely used in several face recognition systems thanks to their wonderful robustness and robust discriminative power. However, most existing binary face descriptors are hand-crafted, that require strong previous knowledge to engineer them by hand. In this paper, we have a tendency to propose a compact binary face descriptor (CBFD) feature learning technique for face representation and recognition. Given every face image, we have a tendency to initial extract pixel difference vectors (PDVs) in local patches by computing the difference between every pixel and its neighboring pixels. Then, we learn a feature mapping to project these pixel difference vectors into low-dimensional binary vectors in an unsupervised manner, where one) the variance of all binary codes in the training set is maximized, 2) the loss between the initial real-valued codes and therefore the learned binary codes is minimized, and three) binary codes evenly distribute at every learned bin, thus that the redundancy data in PDVs is removed and compact binary codes are obtained. Lastly, we tend to cluster and pool these binary codes into a histogram feature as the ultimate representation for each face image. Moreover, we propose a coupled CBFD (C-CBFD) technique by reducing the modality gap of heterogeneous faces at the feature level to make our methodology applicable to heterogeneous face recognition. Extensive experimental results on 5 widely used face datasets show that our methods outperform state-of-the-art face descriptors.
Did you like this research project?
To get this research project Guidelines, Training and Code... Click Here