基于特征融合与嵌入的人脸图像盲修复算法

Blind Face Restoration Algorithm Based on Feature Fusion and Embedding

人脸图像盲修复是从未知退化中恢复出高质量的人脸图像,其不适定性往往会造成修复出的图像出现局部纹理缺失或面部成分不匹配的结果,为此提出基于特征融合与嵌入的人脸图像盲修复算法。通过提取退化输入的面部先验特征,采用多头交叉注意力进行特征交互融合和全局上下文建模,将面部先验嵌入预训练生成网络的潜在空间中,并基于损失函数进行优化,修复因退化而丢失或损坏的局部纹理,实现真实性与忠实度之间的平衡。数值实验在3个真实退化图像数据集上进行,本文方法在客观指标和主观质量上都优于现有方法,最后的消融实验验证了退化人脸图像盲修复算法的有效性。

Blind face restoration is to recover high quality face from unknown degradation,and the illposed problem often results in local texture missing or mismatched facial components for restored images,therefore a degraded blind face restoration algorithm based on feature fusion and embedding optimization is proposed.By extracting face prior features from degraded inputs,using multi-headed cross-attention for feature interaction fusion and global context modeling,embedding facial priors into the latent space of pretrained generative networks,and carrying out optimization based on loss functions,local textures lost or damaged due to degradation are repaired to achieve a balance between realism and fidelity.Numerical experiments are conducted on three real degraded datasets,which outperform existing methods in terms of objective metrics and subjective quality,and the final ablation experiments validate the effectiveness of the degraded blind face restoration algorithm.