基于注意力机制的单幅雨滴图像增强  

SINGLE RAINDROP IMAGE ENHANCEMENT BASED ON ATTENTION MECHANISM

作　　者：郑顾平[1] 李金华 曹锦纲 Zheng Guping;Li Jinhua;Cao Jingang(School of Control and Computer Engineering,North China Electric Power University,Baoding 071003,Hebei,China)

机构地区：[1]华北电力大学控制与计算机工程学院,河北保定071003

出　　处：《计算机应用与软件》2020年第9期160-164,共5页Computer Applications and Software

基　　金：中央高校基本科研业务费专项资金项目(2018MS072)。

摘　　要：在有雨环境下采集到的图片往往会掺杂雨滴或雨条纹,导致图片模糊,而现有去除雨滴实现图像增强算法容易引入更多噪声或不能去除较大雨滴。针对这一问题,提出一种基于注意力机制的单幅雨滴图像增强模型。采用多尺度扩张卷积进行特征提取,引入注意力机制来关注图像中雨滴区域的细节。实验结果表明,该模型可以有效地去除图片上的雨滴,实现雨滴模糊类图像细节的恢复。Pictures collected in a rainy environment are often doped with raindrops or rain stripes,resulting in blurred images.However,the existing image enhancement algorithm for raindrop removal is easy to introduce more noise or fail to remove larger raindrops.Aiming at this problem,this paper proposes a single raindrop image enhancement model based on attention mechanism.It used multi-scale dilated convolution for feature extraction and introduced attention mechanism to pay attention to the details of the raindrop region in the image.The experimental results show that the proposed model can effectively remove the raindrops on the pictures and realize the restoration of the details of the blurred image of raindrops.

关 键 词：扩张卷积 注意力机制 图像增强 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]