| Face recognition performance degrades when face images are of very low resolution since many details about the difference between one person and another can only be captured in images of sufficient resolution. In this work, we propose a new procedure for recognition of low-resolution probe face images, when there is a high-resolution training set available. Most previous super-resolution approaches are aimed at reconstruction, with recognition only as an after-thought. In contrast, in the proposed method, face features, as they would be extracted for a face recognition algorithm (e.g., eigenfaces, Fisherfaces, etc.), are included explicitly in a super-resolution method as prior information. This approach simultaneously provides measures of fit of the super-resolution result, from both reconstruction and recognition perspectives. This is different from the conventional paradigms of matching in a low-resolution domain, or, alternatively, applying a super-resolution algorithm to a low-resolution face image and then classifying the super-resolution result. We show, for example, that recognition of face images of as low as 6 x 6 pixel size is considerably improved compared to matching using a super-resolution reconstruction followed by classification, and to matching with a low-resolution training set.;Simultaneous super-resolution and recognition is a new framework that gives super-resolution the objective of recognition, rather than just reconstruction. Our formulation can be easily expanded or generalized. It is not limited to face or even biometrics, and it is not restricted to a particular super-resolution method or feature extraction algorithm. As demonstrated in this work, the proposed algorithm incorporates models used in super-resolution together with recognition features by including them in a regularization framework. By finding a high-resolution template that fits simultaneously into the available models and features under an assumed class membership, we can extract new features for recognition. Our results show that simple linear discriminants using these features produce better recognition performance than standard approaches. |
