摘要
针对目前图像去雾算法存在特征提取不充分、不能差异化处理导致算法去雾效果差和颜色失真等问题,提出一种基于自校准卷积与特征注意力的图像去雾算法。首先采用自校准卷积模块替代传统卷积模块,扩大网络的感受野并丰富网络的输出特征;其次引入特征注意力模块,对不同通道的所有权重和不同图像像素上分布不均匀的雾度进行差异化处理改善去雾效果,最后结合局部与全局残差学习,进一步提升网络的训练效果。实验结果表明,该算法处理后的图像峰值信噪比(PSNR)和结构相似性(SSIM)的平均值分别为29.799dB和0.967,与对比算法相比分别提高12.9%和3.4%。
Aiming at the problems of insufficient feature extraction and inability to differentiate processing in current image dehazing algorithms,resulting in poor dehazing effect and color distortion,an image dehazing algorithm based on self-calibration convolution and feature attention is proposed.Firstly,the self-calibration convolution module is used to replace the traditional convolution module to expand the receptive field of the network and enrich the output features of the network;Secondly,the feature attention module is introduced to analyze all the weights of different channels and the uneven distribution of haze on different image pixels.Differentiated processing to improve the dehazing effect,and finally combined with local and global residual learning to further improve the training effect of the network.The experimental results show that PSNR and SSIM of the images processed by the algorithm are 29.799dB and 0.967,respectively,which are 12.9%and 3.4%higher than those of the comparison algorithms,respectively.
作者
杨忆
何涛
徐鹤
许广峰
YANG Yi;HE Tao;XU He;XU Guangfeng(College of Electronic and Optical Engineering,College of Flexible Electronics(Future Technology),Nanjing University of Posts and Telecommunications,Nanjing Jiangsu 210023;College of Computer,College of Software,College of Cyberspace Security,Nanjing University of Posts and Telecommunications,Nanjing Jiangsu 210023;School Jiangsu Key Laboratory of Wireless Sensor Network High Technology Research,Nanjing University of Posts and Telecommunications,Nanjing Jiangsu 210003)
出处
《软件》
2022年第7期53-58,62,共7页
Software
基金
国家重点研发计划项目(2019YFB2103003)。
关键词
图像去雾
自校准卷积
特征注意力
残差网络
深度学习
image dehazing
self-calibrating convolution
feature attention
residual network
deep learning
