PROJECT TITLE :

High-resolution face verification using pore-scale facial features - 2015

ABSTRACT:

Face recognition strategies, which sometimes represent face images using holistic or native facial features, rely heavily on alignment. Their performances conjointly suffer a severe degradation below variations in expressions or poses, particularly when there is one gallery per subject only. With the simple access to high-resolution (HR) face images these days, some HR face databases have recently been developed. But, few studies have tackled the employment of HR information for face recognition or verification. During this paper, we propose a create-invariant face-verification method, which is robust to alignment errors, using the HR information based on pore-scale facial features. A replacement keypoint descriptor, namely, pore-Principal Element Analysis (PCA)-Scale Invariant Feature Remodel (PPCASIFT)-custom-made from PCA-SIFT-is devised for the extraction of a compact set of distinctive pore-scale facial options. Having matched the pore-scale features of 2-face regions, a good sturdy-fitting scheme is proposed for the face-verification task. Experiments show that, with one frontal-view gallery only per subject, our proposed method outperforms a variety of normal verification methods, and can achieve wonderful accuracy even the faces are below giant variations in expression and create.