Locality-Constrained Collaborative Representation with Multi-resolution Dictionary for Face Recognition.

Sparse learning methods have drawn considerable attention in face recognition, and there are still some problems need to be further studied. For example, most of the conventional sparse learning methods concentrate only on a single resolution, which neglects the fact that the resolutions of real-world face images are variable when they are captured by different cameras. Although the multi-resolution dictionary learning (MRDL) method considers the problem of image resolution, it takes a lot of training time to learn a concise and reliable dictionary and neglects the local relationship of data. To overcome the above problems, we propose a locality-constrained collaborative representation with multi-resolution dictionary (LCCR-MRD) method for face recognition. First, we extend the traditional collaborative representation based classification (CRC) method to the multi-resolution dictionary case without dictionary learning. Second, the locality relationship characterized by the distance between test sample and training sample is used to learn weight of representation coefficient, and the similar sample is forced to make more contribution to representation. Last, LCCR-MRD has a closed-form solution, which makes it simple. Experiments on five widely-used face databases demonstrate that LCCR-MRD outperforms many state-of-art sparse learning methods. The Matlab codes of LCCR-MRD are publicly available at https://github.com/masterliuhzen/LCCR-MRD.