Single Image Dehazing Based On Convolution Neural Networks

In recent years,with the changes in the environment,haze weather frequently appear,which have a serious impact on vision.There are a lot of dry particles in the air,which will weaken the reflected light by scattering and make the outdoor images degraded.And haze weather may lead serious color distortion,greatly reducing the application value of the image,which results in outdoor photography systems can not work.Therefore,it is necessary to do some research on how to effectively recover the degraded images.In this paper,a large number of literatures are reviewed,and the fog sky imaging model is introduced in detail.The image dehazing algorithm based on enhancement and restoration are systematically expounded.And then focus on the study of learning-based image dehazing algorithm.the ideology,advantages and disadvantages of them are introduced too.the subject of this paper is proposed:single image dehazing algorithm based on convolution neural network.In this paper,single image dehazing based on convolution neural network is described in detail.The main idea is as follows:We propose a deep learning method for single image haze removal.Our method directly learns an end-to-end mapping between the haze images and their corresponding haze layers(the residual images between haze images and non-haze iamges).The mapping is represented as a deep convolutional neural network(CNN)that takes the haze image as the input and outputs the reidual one.and then the revovered haze-free image can be gotten by removing the residual one directly from the fog image without any-other paramaters etismation.Residual learning allows the network to directly estimate the initial haze layer with relatively high learning rates,reducing computational complexity and speeding up the convergence process.Otherwise,unlike traditional dehaze methods that handle each component separately,our method jointly optimizes all layers while training.Since the initial haze layer is rough,we use guided image filter to refine,avoiding halos and block artifacts,which make the recovered is more similar to the real scene.Finally,we implement the proposed algorithm under the framework of Caffe,and then analyze and contrast the experimental results carefully.In this paper,the effect on fog images with different fog density is tested,and many comparisons are listed with other classical algorithms.Experiments demonstrate that the proposed algorithm outperforms state-of-the-art methods on both synthetic and real-world images,qualitatively and quantitatively.