Multidimensional scaling for matching low-resolution facial images

Face recognition performance degrades considerably when the input images are of poor resolution as is often the case for images taken by surveillance cameras or from a large distance. In this paper, we propose a novel approach for the recognition of low resolution images using multidimensional scaling. From a resolution point of view, the scenario yielding the best performance is when both the probe and gallery images are of high enough resolution to discriminate across different subjects. The proposed method embeds the low resolution images in an Euclidean space such that the distances between them in the transformed space approximates the best distances had both the images been of high resolution. The mapping is learned from high resolution training images and their corresponding low resolution images using iterative majorization algorithm. Extensive evaluation of the proposed approach on different datasets like PIE and FRGC with resolution as low as 7 × 6 pixels illustrates the usefulness of the method. We show that the proposed approach significantly improves the matching performance as compared to performing standard matching in the low-resolution domain. Performance comparison with different super-resolution techniques which obtains higher-resolution images prior to recognition further signifies the effectiveness of our approach.