Statistical Learning Based No Reference Face Image Quality Assessment

Face image quality have a significant impact on face detection, face recognition and other face-image-based system's performance. Face image quality assessment differs from the traditional quality assessment. In this paper, the research work about the face image quality assessment can be summarized as follows:(1) The face image quality standard is formulated, and according to this standard, the face images is demarcated strictly based on their quality by ten volunteers, and the average is taken as the final result;(2) Modification, optimization and parallelization on LIBSVM algorithm is researched, which can reduce the training and testing time significantly.(3) With Gabor wavelet feature and the improved LIBSVM algorithm, the classifiers are trained and a binary decision tree is applied in the multi-class face image quality assessment experiments;(4) Classifier fusion method applied on face image quality assessment is researched, which can reduce the classification error and enhance classifiers' capability.The method, which is used for no reference facial image quality assessment in this article, has a good guide on the no reference image quality assessment and lays a good foundation for further research.