关键词: 机器视觉, 人脸超分辨率, 3D先验, 注意力机制

Abstract: In order to effectively solve  the problem of facial super-resolution feature recovery in complex environments, we proposed a novel facial super-resolution network. By integrating 3D rendering prior knowledge and a dual attention mechanism, the network enhanced the understanding of the facial spatial position and overall structure while improving the ability to recover detailed information. The experimental results on the CelebAMask-HQ dataset show that  the proposed algorithm achieves peak signal-to-noise ratio and  structural similarity of 28.76 dB  and  0.827 5 for  downsampled faces magnified by 4 times, and  26.29 dB and 0.754 9 for downsampled faces magnified by 8 times.   Compared with the similar SAM3D algorithm, the proposed algorithm improves the peak signal-to-noise ratio and  structural similarity by  4.09 and 1.93 percentage points when dealing with  4 times  downsampling, and by 2.02 and 4.54 percentage points  when dealing with 8 times downsampling, respectively.  This proves the superiority of the proposed  algorithm and  also indicates that  facial super-resolution recovery can achieve more realistic and clear visual effects in practical applications.

Key words: machine vision, facial super-resolution, 3D prior, attention mechanism