DFQA: Deep Face Image Quality Assessment

Abstract

A face image with high quality can be extracted dependable features for further evaluation, however, the one with low quality can’t. Different from the quality assessment algorithms for general images, the face image quality assessment need to consider more practical factors that directly affect the accuracy of face recognition, face verifcation, etc. In this paper, we present a two-stream convolutional neural network (CNN) named Deep Face Quality Assessment (DFQA) specifically for face image quality assessment. DFQA is able to predict the quality score of an input face image quickly and accurately. Specifcally, we design a network with two-stream for increasing the diversity and improving the accuracy of evaluation. Compared with other CNN network architectures and quality assessment methods for similar tasks, our model is smaller in size and faster in speed. In addition, we build a new dataset containing 3000 face images manually marked with objective quality scores. Experiments show that the performance of face recognition is improved by introducing our face image quality assessment algorithm.