Unconstrained face verification assisted by pairwise visual pre-estimation on key facial points

Investigating that some face regions are possibly more reliable than the others when verifying two face images due to the local abnormal differences caused by the uncontrolled factors in unconstraint environment,we propose a novel face verification algorithm based on pairwise pre-estimation. In our algorithm, we estimate the reliability of a face region by detecting abnormal differences on some key facial points of a face image pair. Then we implement classifications on such reliable regions and combine the results to generate the final recognition result. Furthermore, in the classification, we also propose a pairwise representation based on multiple descriptors and similarity metrics to describe an image pair and train binary-class SVM(Support Vector Machine) classifiers. The extensive experiments on the challenging face database LFW(Labeled Face in the wild) confirm the effectiveness of our method.