Researches On Image Dehazing And Enhancement

Haze is a traditional atmospheric phenomenon where dust,smoke and other dry particles obscure the clarity of the atmosphere.Haze causes issues in the area of terrestrial photography,where the light penetration of dense atmosphere may be necessary to image distant subjects.This results in the visual effect of a loss of contrast in the subject,due to the effect of light scattering through the haze particles.For these reasons,haze removal is desired in both consumer photography and computer vision applications.This paper focuses on image/video dehazing and enhacement,and the main contributions are as follows:1)The key to image dehazing is to estimate the medium transmission.In this paper,a trainable end-to-end system called DehazeNet is proposed to estimate the medium transmission map that is subsequently used to recover a haze-free image via atmospheric scattering model.DehazeNet adopts convolutional neural networks,whose layers are specially designed to embody the established dehazing priors.Specifically,layers of Maxout units are used for feature extraction,and Bilateral Rectified Linear Unit(BReLU)is proposed to improve recovering quality.Experiments show that DehazeNet achieves superior performance over existing methods,yet keeps efficient and easy to use.2)Video dehazing has a more wide range of real-time applications,but additional challenges mainly come from spatio-temporal coherence and computational efficiency.In this paper,a spatio-temporal Markov random field is built with an intensity value prior for real-time video dehazing.Moreover,to facilitate real-time applications,integral image and down sampling technique are approximated to reduce the main computational burden.Experimental results demonstrate that the proposed method is effectively to remove haze and flickering artifacts,and sufficiently fast for real-time applications.3)In atmospheric scattering model,non-gray airlight will result color distortion.When the illumination is decomposed in the RGB-color space,the reflectance retains the original color information of the object,meaning that the Retinex decomposition has the effect to correct color distortion.In this paper,a joint intrinsic-extrinsic prior model is proposed to estimate both illumination and reflectance.The 2D image formed from 3D object in the scene is affected by the intrinsic properties(shape and texture)and the extrinsic property(illumination).Better than conventional Retinex models,the proposed model can preserve the structure information by shape prior,estimate the reflectance with fine details by texture prior,and capture the luminous source by illumination prior.4)Heavy haze results contrast decreasing and detail losing,which can be improved by multi-scale detail enhancement based on scale-aware smoothing.Both in detail and structure regions,the gradient magnitudes are high enough to result edge-aware filters preserving them naturally.This paper presents a novel scale-aware image smoothing via Relativity-of-Gaussian(RoG).As a simple local measure,RoG performs the local analysis of scale features and globally optimizes its results into a piecewise smooth.In particular,a separable recursive optimization is introduced to improve the computational complexity to O(N)and achieve a fast speed.