Double layer coupled locality preserving mappings for very low-resolution face recognition

Abstract

The existing methods based on coupled spatial mappings mostly use high-resolution images and low-resolution images to learn the coupled space for identifying operations. These methods only use high-resolution images of one scale. It is difficult to learn the best coupled space, because there is a big gap between the dimensions of the high-resolution images and the very low-resolution images. To solve this issue, we propose a double layer coupled locality preserving mappings method. Two types of high-resolution template images are used to learn two coupled spaces with preserving the local structure. And the new multi-space fusion similarity measure method is constructed by using two different coupled spaces. The identification task is completed based on the fusion similarity measurement method. In this paper, the more feature information can be used in the training process, and the proposed fusion similarity measurement method can effectively complement feature information and is a more accurate measurement method. The experimental results of the proposed method on three public available face datasets show that the proposed double layer coupled locality preserving spatial mappings method is superior to the state of very low-resolution image recognition methods.