Multi-Scale Residual Channel Attention Network for Face Super-Resolution

To address the problem of low efficiency and reconstruction distortion in current face super-resolution algorithms,a multi-scale residual channel attention network(MSRCAN)is proposed.The network can simultaneously process different upscale factors with a multi-scale progressive structure.Meanwhile,in order to reduce the impact of redundant and invalid features on the network,channel attention mechanism is introduced in the feature reconstruction module of the network,and a novel residual channel attention block is proposed based on face parsing maps,which not only improves the utilization rate of network features but also strengthens the constraints of facial priori.The proposed algorithm is compared with other algorithms in Helen,CelebA and LFW datasets.Extensive experiments show that the proposed algorithm is superior to other existing algorithms,both in subjective visual quality and objective evaluation index such as peak signal-to-noise ratio(PSNR)and structural similarity index(SSIM).