基于条件生成对抗网络的图像去雾算法
|
摘要
为了提高雾天图像的去雾效果,提出了一种基于条件生成对抗网络的图像去雾算法.通过端到端可训练的神经网络对合成的室内和室外数据集进行训练,为了捕捉图像中更多的有用信息,在生成网络中设计了生成器和判别器架构,利用预训练的视觉几何组特征模型和L_1-正则化梯度对损失函数进行修正,并在判别器的最后一层引入Sigmoid函数用于特征映射,以便进行概率分析可归一化.利用合成数据集对损失函数进行训练,得到新的损失函数的参数,然后利用室外自然有雾图像数据集对训练得到的新的损失函数进行测试.实验结果表明:所提算法有效解决了去雾图像的颜色失真、过饱和、视觉伪像等问题,生成效果更好的去雾图像.
In order to improve the dehazing effect of foggy images,an image defogging algorithm based on conditional generation against network was proposed.Through end-to-end trainable nerves,the network trained the synthesized indoor and outdoor data sets.In order to capture more useful information in the image,the generator and discriminator architecture was designed in the generation network,the loss function was modified using the pre-trained visual geometry group feature model and the L_1-regular gradient pair loss.At the last level of the discriminator,the Sigmoid function was applied to the feature map for probabilistic analysis to be normalized.By using the synthetic data set to train the loss function,the parameters of the new loss function were obtained,and then the new trained loss function was tested by the outdoor natural fog image data set.The experimental results show that the algorithm effectively solves the problem of color distortion,oversaturated,and visual artifacts,resulting in a better defogging image.
In order to improve the dehazing effect of foggy images,an image defogging algorithm based on conditional generation against network was proposed.Through end-to-end trainable nerves,the network trained the synthesized indoor and outdoor data sets.In order to capture more useful information in the image,the generator and discriminator architecture was designed in the generation network,the loss function was modified using the pre-trained visual geometry group feature model and the L_1-regular gradient pair loss.At the last level of the discriminator,the Sigmoid function was applied to the feature map for probabilistic analysis to be normalized.By using the synthetic data set to train the loss function,the parameters of the new loss function were obtained,and then the new trained loss function was tested by the outdoor natural fog image data set.The experimental results show that the algorithm effectively solves the problem of color distortion,oversaturated,and visual artifacts,resulting in a better defogging image.
引文
[1] YANG Hong,CUI Yan.Image dehazing algorithm based on open computing dark channel and optimizing boundary constraint[J].Acta Photonica Sinica,2018,47(6):0610004.杨红,崔艳.基于开运算暗通道和优化边界约束的图像去雾算法[J].光子学报,2018,47(6):0610004.
[2] HE K,SUN J,TANG X.Single image haze removal using dark channel prior[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2011,33(12):2341-2353.
[3] GALDRAN A,VAZQUEZ C,PARDO D,et al.Enhanced variational image dehazing[J].Society for Industrial and Applied Mathematics Journal on Imaging Sciences,2015,8(3):1519-1546.
[4] NISHINO K,KRATZ L,LOMBARDI S.Bayesian defogging[J].International Journal of Computer Vision,2012,98(3):263-278.
[5] FATTALR.Dehazing using color-lines[J].ACM Transactions on Graphics,2014,34(1):13.
[6] TANG K.YANG J,WANGJ.Investigating haze-relevant features in a learning framework for image dehazing[C].Conference on Computer Vision and Pattern Recognition,2014:2995-3002.
[7] CAI B,XU X,JIA K,et al.Dehazenet:an end-to-end system for single image haze removal[J].IEEE Transactions on Image Processing,2016,25(11):5187-5198.
[8] REN W,LIU S,ZHANG H,et al.Single image dehazing via multiscale convolutional neural networks[C].European Conference on Computer Vision,2016:154-169.
[9] LI B,PENG X,WANGZ,et al.Aod-net:allin-one dehazing network[C].International Conference on Computer Vision,2017:4780-4788.
[10] GOODFELLOW I,POUGET A J,MIRZAM,et al.Generative adversarial nets[C].International Conference on Neural Information Processing Systems,2014:2672-2680.
[11] SOHN K,LEE H,YAN X.Learning structured output representation using deep conditional generative models[C].International Conference on Neural Information Processing Systems,2015:3483-3491.
[12] YEH R A,CHEN C,YIANL T,et al.Semantic image inpainting with deep generative models[C].Conference on Computer Vision and Pattern Recognition,2017:6882-6890.
[13] REED S,AKATA Z,YANX,et al.Generative adversarial text to image synthesis[C].International Conference on International Conference on Machine Learning,2016:1060-1069.
[14] LEDIG C,THEIS T,HUSZR F,et al.Photo-realistic single image super-resolution using agenerative adversarial network[C].Conference on Computer Vision and Pattern Recognition,2017:105-114.
[15] ZHANG H,SINDAGI V,PATELV M.Image de-raining using a conditional generative adversarial network[J].arXiv,2017:1701.05957.
[16] HE K,ZHANG X,REN S,et al.Deep residual learning for image recognition[C].Conference on Computer Vision and Pattern Recognition,2016:770-778.
[17] RONNEBERGER O,FISCHER P,BROX T.Unet:convolutional networks for biomedical image segmentation[C].Medical Image Computing and Computer-Assisted Intervention,2015:234-241.
[18] KINGMA D P,BA J L.Adam:a method for stochastic optimization[J].arXiv,2017:1412.6980.
[2] HE K,SUN J,TANG X.Single image haze removal using dark channel prior[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2011,33(12):2341-2353.
[3] GALDRAN A,VAZQUEZ C,PARDO D,et al.Enhanced variational image dehazing[J].Society for Industrial and Applied Mathematics Journal on Imaging Sciences,2015,8(3):1519-1546.
[4] NISHINO K,KRATZ L,LOMBARDI S.Bayesian defogging[J].International Journal of Computer Vision,2012,98(3):263-278.
[5] FATTALR.Dehazing using color-lines[J].ACM Transactions on Graphics,2014,34(1):13.
[6] TANG K.YANG J,WANGJ.Investigating haze-relevant features in a learning framework for image dehazing[C].Conference on Computer Vision and Pattern Recognition,2014:2995-3002.
[7] CAI B,XU X,JIA K,et al.Dehazenet:an end-to-end system for single image haze removal[J].IEEE Transactions on Image Processing,2016,25(11):5187-5198.
[8] REN W,LIU S,ZHANG H,et al.Single image dehazing via multiscale convolutional neural networks[C].European Conference on Computer Vision,2016:154-169.
[9] LI B,PENG X,WANGZ,et al.Aod-net:allin-one dehazing network[C].International Conference on Computer Vision,2017:4780-4788.
[10] GOODFELLOW I,POUGET A J,MIRZAM,et al.Generative adversarial nets[C].International Conference on Neural Information Processing Systems,2014:2672-2680.
[11] SOHN K,LEE H,YAN X.Learning structured output representation using deep conditional generative models[C].International Conference on Neural Information Processing Systems,2015:3483-3491.
[12] YEH R A,CHEN C,YIANL T,et al.Semantic image inpainting with deep generative models[C].Conference on Computer Vision and Pattern Recognition,2017:6882-6890.
[13] REED S,AKATA Z,YANX,et al.Generative adversarial text to image synthesis[C].International Conference on International Conference on Machine Learning,2016:1060-1069.
[14] LEDIG C,THEIS T,HUSZR F,et al.Photo-realistic single image super-resolution using agenerative adversarial network[C].Conference on Computer Vision and Pattern Recognition,2017:105-114.
[15] ZHANG H,SINDAGI V,PATELV M.Image de-raining using a conditional generative adversarial network[J].arXiv,2017:1701.05957.
[16] HE K,ZHANG X,REN S,et al.Deep residual learning for image recognition[C].Conference on Computer Vision and Pattern Recognition,2016:770-778.
[17] RONNEBERGER O,FISCHER P,BROX T.Unet:convolutional networks for biomedical image segmentation[C].Medical Image Computing and Computer-Assisted Intervention,2015:234-241.
[18] KINGMA D P,BA J L.Adam:a method for stochastic optimization[J].arXiv,2017:1412.6980.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |