Face image quality assessment for face selection in surveillance video using convolutional neural networks

Automated Face Quality Assessment (FQA) plays a key role in improving face recognition accuracy and increasing computational efficiency. In the context of video, it is very common to acquire multiple face images of a single person. If one were to use all the acquired face images for the recognition task, the computational load for Face Recognition (FR) increases while recognition accuracy decreases due to outliers. This impediment necessitates a strategy to optimally choose the good quality face images from the pool of images in order to improve the performance of the FR algorithm. Toward this end, we propose a FQA algorithm that is based on mimicking the recognition capability of a given FR algorithm using a Convolutional Neural Network (CNN). In this way, we select those face images that are of high quality with respect to the FR algorithm. The proposed algorithm is simple and can be used in conjunction with any FR algorithm. Preliminary results demonstrate that the proposed method is on par with the state-of-the-art FQA methods in improving the performance of FR algorithms in a surveillance scenario.