The Research Of Image Dehazing Algorithm Based On Atmospheric Scattering Model

In recent years,haze has become a normal weather condition in our daily life,leading to the degradation of contrast and color fidelity of outdoor images.This phenomenon not only affects the visual beauty,but also disturbs the image-based follow-up works,such as image retrieval and classification,the target reconnaissance under the monitoring scene,and the identification and tracking of the target in the intelligent system.Therefore,image dehazing plays an important role in practical applications.How to obtain high quality recovered image from a single image is a challenging problem in dehazing research,which has drawn more and more scholars' attention.In this thesis,single image haze removal algorithms are studied on the basis of the atmospheric scattering model.In view of the limitations of current dehazing algorithms,we focus on proposing a scheme with lower complexity,more accurate estimation of transmission and better visibility of the dehazed images.The main innovative work includes:(1)Aiming at the problem of high time complexity and color distortion in the bright regions of recovered images occurred in dark channel prior algorithm,this thesis proposes a fast haze removal method using an adaptive compensation technique based on spatial location.Instead of refining the transmission map with softmatting or other operations to reduce the complexity greatly,our method estimates the initial transmission using pixel-level transmission which is able to preserve the edge information.For the case that the under-estimated transmission in the bright regions which are not satisfied with the dark channel prior,an adaptive function that combines with the spatial location is presented to compensate the initial transmission,and a gaussian filter is used to smooth the texture noise.The comparison against several existing dehazing methods in terms of dehazing effect and efficiency shows that the transmission obtained by our method is more accurate,and the restored image is more natural,especially for the regions that the dark channel prior is invalid.Furthermore,this method has a low computational consumption.Compared with the original dehazing algorithm based on dark channel prior,the proposed method performs hundreds of times faster.Moreover,the computational time of the proposed method reduces about four times than the current fast dehazing algorithm.(2)Aiming at the problem that the existing dehazing algorithms do not consider the relationship between the edge and scene depth which induces misestimation of the transmission,this thesis proposes a weighted fusion haze removal approach based on edge classification.Firstly,we put forward a novel prior,named as depth edge prior,and then design an edge classifier to acquire the depth edges which can reflect the depth changes.On this basis,an effective transmission estimation method is developed by using a pixel-to-patch fusion strategy weighted by depth edges,and the transmission in the regions which are not satisfied with the dark channel prior is compensated according to the histogram distribution of the transmission.Experiments on a number of haze removal algorithms verify that proposed method can produce a more accurate and reasonable transmission by preserving the sharp discontinuities at the location of the depth edges and keeping the smoothness in rest regions.The restored images obviously have higher contrast,more details,and less distortion.