Severe weather conditions such as rain and snow adversely affect the visualquality of images captured under such conditions thus rendering them uselessfor further usage and sharing. In addition, such degraded images drasticallyaffect performance of vision systems. Hence, it is important to solve theproblem of single image de-raining/de-snowing. However, this is a difficultproblem to solve due to its inherent ill-posed nature. Existing approachesattempt to introduce prior information to convert it into a well-posed problem.In this paper, we investigate a new point of view in addressing the singleimage de-raining problem. Instead of focusing only on deciding what is a goodprior or a good framework to achieve good quantitative and qualitativeperformance, we also ensure that the de-rained image itself does not degradethe performance of a given computer vision algorithm such as detection andclassification. In other words, the de-rained result should beindistinguishable from its corresponding clear image to a given discriminator.This criterion can be directly incorporated into the optimization framework byusing the recently introduced conditional generative adversarial networks(GANs). To minimize artifacts introduced by GANs and ensure better visualquality, a new refined loss function is introduced. Based on this, we propose anovel single image de-raining method called Image De-raining ConditionalGeneral Adversarial Network (ID-CGAN), which considers quantitative, visual andalso discriminative performance into the objective function. Experimentsevaluated on synthetic images and real images show that the proposed methodoutperforms many recent state-of-the-art single image de-raining methods interms of quantitative and visual performance.
展开▼
