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

Abstract:

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.

Key words: image super-resolution, Swin Transformer, deep learning, conditional priori, face super-resolution, attention mechanism, Transformer, image processing, super reconstruction