Robust Face Recognition via Multi-scale Contextual Information Ensemble Learning

The study of robust face recognition has always gained much attention and has been widely applied to many social and public safety protection fields. Image patch based methods have achieved much more attractive performance. Especially, multi-scale patch based methods take the impact of different scales of the image patch into consideration. However, these methods ignore a fact that different parts of the face image contain different contextual semantic information which may be useful for the recognition task. To this end, in this study, we presented a robust face recognition approach by fully utilizing the multi-scale contextual information. Different from previous patch based methods, in our method, we select several patches around the test patch with a certain step in each window. Then the concatenated patch set is linearly represented over the corresponding patch sets on the training samples. By this consideration, the contextual topology can provide complementary contributions to the recognition, especially when the test faces have occlusions. Also, the multi-scale ensemble learning scheme is exploited to further enhance the recognition performance. Our extensive experiments have validated the superiority of our proposed approach over some state-of-the-art patch based robust face recognition approaches.