Abstract

HERNANDEZ-DURAN, Mairelys  and  PLASENCIA-CALANA, Yenisel. Metric Learning  for Low-Resolution Face Recognition. Rev cuba cienc informat [online]. 2016, vol.10, n.1, pp.124-133. ISSN 2227-1899.

ABSTRACT Low-resolution face recognition is a very difficult problem. In this setup, the database or gallery contains high-resolution images but the image to be recognized is a low-resolution one. Thus we are dealing with a resolution mismatch problem for training and test images. Standard face recognition methods fail in this setting, which suggests that current feature representation approaches are not adequate to cope with this problem. Therefore, we propose the use of dissimilarity representations as an alternative to the use of feature representations. This work is an extension of a previous one, in which the dissimilarity space was used for low-resolution face recognition. In this paper we propose to replace a Euclidean distance computed over the vector features for a Mahalanobis distance, which is a metric automatically learned by optimizing a classification criterion in the training set. We also propose to replace the Euclidean distance in the dissimilarity space by a metric automatically learned. Experiments on two standard face datasets demonstrate that the use of metric learning outperforms the initial Euclidean distance for low-resolution face recognition. To solve the mismatch problem, the best strategy obtained in previous work was used, which consist on subsample and then scale the training images and scale test.

Keywords : dissimilarity space; low-resolution; face recognition; super-resolution; prototype selection.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )