基于条件先验Swin Transformer的人脸图像超分辨重建  

Face Image Super-Resolution Reconstruction Based on Conditional Priori Swin Transformer

作　　者：郑方亮 王延年[1] 廉继红[1] 阮佩 ZHENG Fangliang;WANG Yannian;LIAN Jihong;RUAN Pei(School of Electronics and Information,Xi'an Polytechnic University,Xi'an 710048,China)

机构地区：[1]西安工程大学电子信息学院,陕西西安710048

出　　处：《电子科技》2025年第2期35-41,共7页Electronic Science and Technology

基　　金：陕西省重点研发计划(2021GY-076);西安工程大学(柯桥)研究生创新学院研究生联合培养项目(19KQYB02)。

摘　　要：针对现有基于Swin Transformer图像超分辨模型未对人脸图像进行预处理导致最终超分辨结果不佳的问题,文中提出了基于条件先验Swin Transformer的人脸图像超分辨重建方法。该方法利用人脸解析图融合Swin Transformer模型对人脸图像进行预处理,使用条件先验对人脸超分问题进行优化,采用人脸解析图Parsing Map进行约束从而得到更有价值的先验信息。在深层特征提取阶段,将通道空间注意力机制融合Swin Transformer模块对特征组调整进行速度与精度的平衡。实验结果表明,所提方法在测试集上的峰值信噪比(Peak Signal-to-Noise Ratio,PSNR)为32.21 dB,相较于现有模型具有一定提升。实验证明改进模型更适用于人脸,所生成结果更清晰、更真实,能够还原出更多人脸图像纹理细节。In view of the problem that the existing image super resolution models based on Swin Transformer do not preprocess the face image,resulting in poor final super resolution results,this study proposes a face image super resolution reconstruction method based on conditional prior Swin Transformer.The method uses face Parsing Map and Swin Transformer model to preprocess the face image,uses conditional prior to optimize the face hyper-segmentation problem,and uses face parsing map to restrict the process so as to obtain more valuable prior information.In the stage of deep feature extraction,the channel space attention mechanism is integrated with Swin Transformer module to balance the speed and precision of feature group adjustment.Experimental results show that the proposed method achieves a PSNR(Peak Signal-to-Noise Ratio)of 32.21 dB on the test set.Compared with the existing model,this method has a certain improvement.It is proved that the improved model is more suitable for human face,and the generated result is clearer and more real,and more details of face image texture can be restored.

关 键 词：图像超分辨 Swin Transformer 深度学习 条件先验 人脸超分辨 注意力机制 TRANSFORMER 图像处理 超分重建 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]