Semi-coupled basis and distance metric learning for cross-domain matching: Application to low-resolution face recognition

Abstract

In this paper, we propose a method for matching biometric data from disparate domains. Specifically, we focus on the problem of comparing a low-resolution (LR) image with a high-resolution (HR) one. Existing coupled mapping methods do not fully exploit the HR information or they do not simultaneously use samples from both domains during training. To this end, we propose a method that learns coupled distance metrics in two steps. In addition, we propose to jointly learn two semi-coupled bases that yield optimal representations. In particular, the HR images are used to learn a basis and a distance metric that result in increased class-separation. The LR images are then used to learn a basis and a distance metric that map the LR data to their class-discriminated HR pairs. Finally, the two distance metrics are refined to simultaneously enhance the class-separation of both HR class-discriminated and LR projected images. We illustrate that different distance metric learning approaches can be employed in conjunction with our framework. Experimental results on Multi-PIE and SCface, along with the relevant hypothesis tests, provide evidence of the effectiveness of the proposed approach. Figure 4. Depiction of the ROC curves for the Multi-PIE database for Experiment 2 (color figure, best viewed in electronic format).

Multi-Scale Learning for Low-Resolution Person Re-Identification

“…In the last five years there are several coupled transformation based subspace models developed for LR face recognition [30,29,2,24,43,14]. A basic idea of these works is to learn coupled transformations such that a LR image can directly match a HR image.…”

Designing a Low-Resolution Face Recognition System for Long-Range Surveillance

“…Li et al [3] proposed a method that projects both high-resolution gallery and low-resolution probe to a unified feature space for classification using coupled mappings which minimize the difference between corresponding images. Moutafis and Kakadiaris [4] proposed a method that learns semi-coupled mappings for optimized representations. The mappings aim at increasing class-separation for highresolution images and mapping low-resolution images to their corresponding class-separated high-resolution data.…”

Face Recognition at a Distance

“…However, the high-frequency information, which can be useful in face recognition, is lost in the down-sampling process. To make maximum use of the gallery image information, recently researchers have put more effort into low-resolution-tohigh-resolution comparison to seek for an optimal solution [14,15]. A face recognition system is usually designed for a certain resolution, and it is not guaranteed that it also performs well for other resolutions.…”

“…This method is also suitable for images of different modalities, for example, visible and infrared faces. Moutafis and Kakadiaris [15] proposed a method that learns semi-coupled mappings for high-resolution and low-resolution images for optimized representations. The mappings aim at increasing classseparation for high-resolution images and projecting low-resolution images to 2.4 Low-resolution face alignment 17 their corresponding class-separated high-resolution data.…”

“…This paper follows the approach of [15] and focuses on LR to HR comparison. We propose a novel method for direct comparison of different resolution images which is called Mixed Resolution Biometric Comparison.…”