基于深度学习和小波变换的图像超分辨率方法  

Image Super-Resolution Method Based on Deep Learning and Wavelet Transform

作　　者：黄玉凤 张小波 HUANG Yufeng;ZHANG Xiaobo(Xi'an Shiyou University,Xi'an 710065,China;Xianyang Normal University,Xianyang 712000,China)

机构地区：[1]西安石油大学,陕西西安710065 [2]咸阳师范学院,陕西咸阳712000

出　　处：《现代信息科技》2025年第6期62-66,共5页Modern Information Technology

摘　　要：近年来,基于深度学习的单幅图像超分辨率方法已取得显著成就。然而,许多方法仅研究图像空域,忽略了图像频域中高频信息的重要性,从而导致生成的图像相对平滑。由于小波变换能够提取图像细节的特性,文章提出一种结合小波变换与深度学习的单幅图像超分辨率方法。该方法通过动态卷积处理图像,利用Haar小波将图像分解为低频子图和三个方向上的高频子图,使用残差密集块和残差块相连作为推理网络。再通过小波逆变换和残差密集块增强局部特征,得到超分辨率图像。实验结果表明,该方法在主观视觉效果与客观评价指标上均表现良好。In recent years,single image Super-Resolution methods based on Deep Learning have achieved remarkable achievements.However,many methods only study the image spatial domain,ignoring the importance of high-frequency information in the image frequency domain,resulting in a relatively smooth image.Because wavelet transform can extract image details,this paper proposes a single image Super-Resolution method combining wavelet transform and Deep Learning.The method processes the image through dynamic convolution,uses Haar wavelet to decompose the image into low-frequency subgraphs and high-frequency subgraphs in three directions,and employs connected residual dense blocks and residual blocks to construct the inference network.Then,the local features are enhanced by inverse wavelet transform and residual dense blocks to obtain Super-Resolution images.The experimental results show that the method performs well in both subjective visual effects and objective evaluation indicators.

关 键 词：超分辨率 小波变换 深度学习 残差密集块 残差块 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]