PROJECT TITLE :

Learning Compact Feature Descriptor and Adaptive Matching Framework for Face Recognition - 2015

ABSTRACT:

Dense feature extraction is changing into increasingly common in face recognition tasks. Systems based on this approach have demonstrated spectacular performance in a very vary of challenging situations. But, enhancements in discriminative power return at a computational value and with a risk of over-fitting. During this paper, we tend to propose a replacement approach to dense feature extraction for face recognition, that consists of 2 steps. 1st, an encoding theme is devised that compresses high-dimensional dense options into a compact representation by maximizing the intrauser correlation. Second, we tend to develop an adaptive feature matching algorithm for effective classification. This matching methodology, in contrast to the previous strategies, constructs and chooses a small subset of coaching samples for adaptive matching, ensuing in additional performance gains. Experiments using several difficult face databases, including labeled Faces in the Wild knowledge set, Morph Album 2, CUHK optical-infrared, and FERET, demonstrate that the proposed approach consistently outperforms the current state-of-the-art.