基于改进CGAN网络的图像去雾算法  

Image Dehazing Algorithm Based on Improved CGAN Network

作　　者：程园园 程晓荣[1] CHENG Yuanyuan;CHENG Xiaorong(School of Control and Computer Engineering,North China Electric Power University,Baoding 071000)

机构地区：[1]华北电力大学控制与计算机工程学院,保定071000

出　　处：《计算机与数字工程》2025年第3期845-850,876,共7页Computer & Digital Engineering

摘　　要：为了解决雾天图像与视频的质量大幅度下降的问题,提出了基于改进条件生成对抗网络(CGAN)的图像去雾方法。在传统的生成器中设计添加残差网络模块以及密集空洞空间金字塔池化(DenseASPP)模块来实现多尺度特征的提取,提高特征利用率,增强生成图像的去雾细节保持。判别器使用34×34的PatchGAN进行分块判定,提高图像判别准确度。在合成有雾数据集RESIDE中,通过与暗通道算法、DehazeNet、AOD-Net、传统CGAN算法进行对比,主观上可以看出该网络模型的雾残留少,细节信息的保持和色彩对比度都有所提高。通过峰值信噪比(PSNR)和结构相似度(SSIM)结果对比,客观表明该网络模型恢复无雾图像的效果得到了提升。In order to solve the problem that the quality of foggy images and videos is greatly reduced,an image dehazing method based on an improved Conditional Generative Adversarial Network(CGAN)is proposed.In the traditional generator,the residual network module and the Dense Space Pyramid Pooling(DenseASPP)module are designed and added to achieve multi-scale feature extraction,improve feature utilization,and enhance the preservation of dehazing details of generated images.The discriminator uses 34×34 PatchGAN for block determination,which improves the accuracy of image discrimination.In the synthetic foggy dataset RESIDE,by comparing with the dark channel algorithm,DehazeNet,AOD-Net,and the traditional CGAN algorithm,it can be seen subjectively that the network model has less fog residue,and the preservation of detail information and color contrast are improved.improve.Through the comparison of peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)results,it objectively shows that the effect of this network model in restoring haze-free images has been improved.

关 键 词：单幅图像去雾 条件生成对抗网络 残差网络 DenseASPP PatchGAN 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]