A Novel Deep-Learning-Based Enhanced Texture Transformer Network for Reference Image Super-Resolution

• DOI: 10.3390/electronics11193038
• Publication Date: 2022-09-26T03:13:27Z
• Authors: Changhong Liu; Hongyin Li; Zhongwei Liang; Yongjun Zhang; Yier Yan; Ray Y. Zhong; Shaohu Peng
• Keywords: Feature (linguistics)

The study explored a deep learning image super-resolution approach which is commonly used in face recognition, video perception and other fields. These generative adversarial networks usually have high-frequency texture details. relevant textures of high-resolution images could be transferred as reference to low-resolution images. latest existing methods use transformer ideas transfer related images, but there are still some problems with channel detailed textures. Therefore, the proposed an enhanced network (ETTN) improve ability details texture. It learn corresponding structural information convert it into Through this, finding feature map can change exact between channels. We then multi-scale integration (MSFI) further enhance effect fusion achieved different degrees restoration. experimental results show that model has good resolution enhancement on transformers. In datasets, peak signal noise ratio (PSNR) similarity (SSIM) were improved by 0.1–0.5 dB 0.02, respectively.