%0 Journal Article
%T 基于Transformer和不可分加性小波的图像超分辨率重建<br>Image Super-Resolution Reconstruction Based on Transformer and Non-Separable Additive Wavelet
%A 刘斌
%A 杜丹丹
%J Journal of Image and Signal Processing
%P 40-50
%@ 2325-6745
%D 2023
%I Hans Publishing
%R 10.12677/JISP.2023.121005
%X 针对目前超分辨率重建存在纹理模糊、扭曲等问题，提出了一种基于Transformer和不可分加性小波的网络。该网络由小波分解模块、纹理提取模块、浅层特征提取模块、用于纹理匹配的相关嵌入模块、纹理传输模块、用于纹理融合的跨尺度集成模块共六个模块组成。我们对此网络的重建性能在相关的测试数据上与已有的典型方法进行了对比研究，实验结果表明，该网络不仅提升了视觉效果，而且获得了较好的客观指标评价，在所比较的六种方法中，所构建网络在CUFED5、Sun80和Manga109数据集上的峰值信噪比(PSNR)和结构相似度(SSIM)都具有最佳性能。<br />
Aiming at the problems of texture blur and distortion in super-resolution reconstruction, a network based on Transformer and non-separable additive wavelet is proposed. The network consists of six modules: Wavelet Decomposition module, Texture Extraction module, Shallow Feature Extraction module, Relevance Embedding module for texture matching, Texture Transmission module, and Cross Scale Integration module for texture fusion. We compared the reconstruction performance of this network with the existing typical methods on the relevant test data. The experimental results show that this network not only improves the visual effect, but also obtains better objective index evaluation. Among the six methods compared, the peak signal to noise ratio (PSNR) and structure similarity (SSIM) of the constructed network on CUFED5, Sun80 and Manga109 datasets have the best performance.
%K Transformer，不可分加性小波，超分辨率重建，计算机视觉，深度学习，注意力机制<br>Transformer
%K Non-Separable Additive Wavelet
%K Super-Resolution Reconstruction
%K Computer 
Vision
%K Deep Learning
%K Attention Mechanism
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=60382