Face Quality Measure for Face Authentication

摘要

In a face authentication system, face image quality can significantly influence system performance. Designing an effective image quality measure is necessary to reduce the number of poor quality face images acquired during enrollment and authentication, thereby improving system performance. Furthermore, image quality scores can be used as weights in multimodal system based on weighted score level fusion. In this paper, the authors examined image quality factors, such as contrast, brightness, focus and illumination, and defined quality measure for these factors. The quality measure used template image's, or registration image's, quality as reference quality. Thus, the quality measure does not rely on any reference good quality and criteria to evaluate how good a face image is. The quality measure reflects difference in quality between a template image and a query image. Then, we proposed a face quality measure by combining these factors. Finally, we conducted experiments to evaluate the relationship between face authentication performance and individual image quality factors as well as the combined face quality measure.