Image quality assessment model based on features and applications in face recognition

Facial appearance changes because uncontrolled variations of facial appearances due to illumination, pose, expression, occlusion of non-cooperative subjects and subject-to-camera distance need to be handled to allow for successful recognition. This paper presents a novel image quality assessment model. The model is designed to reduce the influence which is caused by the degradation of facial image quality due to uncontrolled variations of facial appearances, and the degradation can lower the recognition performance. The model assesses the image quality from several aspects: (I) Occlusion measure. (II) Face-to-camera distance measure. (Ill) Pose and expression measure.(IV) Uneven illumination measure. Then noisy score is calculated by the image quality assessment model while higher noisy score´s images will be discarded for face recognition, the superior face images are selected by image quality assessment model to obtain best recognition result. Experimental results on CAS-PEAL face databases with varied uncontrolled facial appearances demonstrated that the proposed approach achieved satisfactory recognition rate.