Patch-based Probabilistic Image Quality Assessment for Face Selection and Improved Video-based Face Recognition

摘要

In video based face recognition, face images are typically captured overmultiple frames in uncontrolled conditions, where head pose, illumination,shadowing, motion blur and focus change over the sequence. Additionally,inaccuracies in face localisation can also introduce scale and alignmentvariations. Using all face images, including images of poor quality, canactually degrade face recognition performance. While one solution it to useonly the "best" subset of images, current face selection techniques areincapable of simultaneously handling all of the abovementioned issues. Wepropose an efficient patch-based face image quality assessment algorithm whichquantifies the similarity of a face image to a probabilistic face model,representing an "ideal" face. Image characteristics that affect recognition aretaken into account, including variations in geometric alignment (shift,rotation and scale), sharpness, head pose and cast shadows. Experiments onFERET and PIE datasets show that the proposed algorithm is able to identifyimages which are simultaneously the most frontal, aligned, sharp and wellilluminated. Further experiments on a new video surveillance dataset (termedChokePoint) show that the proposed method provides better face subsets thanexisting face selection techniques, leading to significant improvements inrecognition accuracy.