Coupled Metric Learning for Face Recognition with Degraded Images

This work formulate the coupled metric learning as an optimization problem and solve it efficiently with a closed-form solution, which can be generally applied to face recognition problems with various degrade images.

Abstract

Real-world face recognition systems are sometimes confronted with degraded face images, e.g., low-resolution, blurred, and noisy ones. Traditional two-step methods have limited performance, due to the disadvantageous issues of inconsistent targets between restoration and recognition, over-dependence on normal face images, and high computational complexity. To avoid these limitations, we propose a novel approach using coupled metric learning, without image restoration or any other preprocessing operations. Different from most previous work, our method takes into consideration both the recognition of the degraded test faces as well as the class-wise feature extraction of the normal faces in training set. We formulate the coupled metric learning as an optimization problem and solve it efficiently with a closed-form solution. This method can be generally applied to face recognition problems with various degrade images. Experimental results on various degraded face recognition problems show the effectiveness and efficiency of our proposed method.