Independent components based blur face identification

To deal with the image blur problem during face recognition, in this paper, the necessity to add a blur identification step ahead of facial image acquiring for recognition is first discussed, and then a components based blur identification approach is proposed. The most significant characteristics for blur identification reside in the high frequency informations of image, and as to a specific facial image, those informations mainly distribute on face components such as eyes, brow, mouth, and so on. Thus we explore to extract features from these face components to exclude the disturbance of other face parts such as cheek and forehead whose dominant informations are contained in low frequency. Specifically, our algorithm relies on the high frequency DCT coefficients on face component as features, then the Random Forest strategy is utilized as the component level identifier to blur, and finally component voting in enrolled to determine the final decision. The effectiveness of the proposed component features and our independent components based blur face identification approach are demonstrated by tremendous experiments on the publicly available FRGC dataset.