基于通道注意力和门控循环单元的图像去雨算法

在计算机视觉领域,雨线或者雨滴会使雨天拍摄的图像变得模糊,降低图像的质量.针对雨天图像质量低下的问题,提出了一种基于通道注意力和门控循环单元的图像去雨算法.该算法基本思路如下:首先将训练图像通过残差记忆模块提取特征;其次将提取的特征通过特征增强模块增加感受野,识别不同等级的雨线特征并将其增强,传递给后续的循环网络;最后网络循环过程中,通过门控循环单元块实现不同循环阶段之间的参数共享.实验结果利用客观评价指标和主观视觉效果进行评估,验证了该算法在较为复杂数据集上的有效性.

Image rain removal algorithm based on channel attention and gated recurrent unit

In the field of computer vision, rain streaks or raindrops may make the images taken on rainy days blurred, and reduce the quality of the image. Aiming at the problem of low image quality in rainy days, this paper proposed a novel image rain removal algorithm based on channel attention and gated recurrent unit to remove rain streaks. The basic idea is as follows: first of all, it extracted features from the training image through the residual memory block; then, increased receptive field with the extracted features through the feature enhancement module, identified and enhanced rain streaks features of different le-vels, which were passed to the subsequent recurrent network; finally, during the recurrent network, the gated recurrent unit block achieved parameter sharing between different recurrent stages. The experimental results verify the effectiveness of the algorithm on complex datasets by using objective evaluation indexes and subjective visual effects, and the work of rain removal is completed.