多尺度残差群网络的图像去雨算法

An Image Deraining Algorithm for Multi-Scale Residual Group Networks

雨天作为一种常见的天气条件会对计算机视觉造成一定影响,图像中会出现雨纹与细节模糊,从而需要一种高效的单幅图像去雨算法来改善图像质量。现有图像去雨算法大多只关注于去除雨纹,而忽略了恢复去雨后图像的细节信息。针对此,为了更好地检测雨纹提出了浅层特征提取模块和深层特征提取模块,其中,浅层特征提取模块选取残差密集块,深层特征提取模块选取两个双注意力模块和两个卷积层作为残差块构成的残差群。为了恢复图像细节信息,提出了一种包含全局分支和局部分支的多尺度细节恢复模块。在合成数据集和真实数据集上的大量实验表明,所提算法的PSNR和SSIM分别达到了40.41 dB和0.989,同时保留了图像细节信息。

As a common weather condition,rainy weather can have a certain impact on computer vision,causing rain streaks and blurred details in images.Therefore,an efficient single image rain removal algorithm is needed to improve image quality.Most existing rain removal algorithms only focus on removing rain streaks,while neglecting the restoration of detailed information in the image afterremoval.In order to better detect rain streaks,a shallow feature extraction module and a deep feature extraction module is proposed.In the shallow extraction module,the residual dense block is selected,while in the deep extraction module,two dual-attention modules and two convolution layers are selected as residual groups composed of residual blocks.In order to restore image detail information,a multi-scale detail restoration module containing global and local branches is proposed.Numerous experiments on both synthetic and real datasets have shown that the proposed algorithm achieves PSNR and SSIM of 40.41 dB and 0.989 respectively,while preserving image details.