Research On Image Dehazing Algorithm Based On Transmission Optimization

Computer vision system has played an increasingly important role in our daily life nowadays.However,as the haze weather seriouslydegrades the images captured by outdoor imaging systems,the contrast of the images has been greatly reduced.The image haze removal technology aims to reduce or eliminate the negative impact of weather conditions such as haze on the imaging system,and to restore the contrast and chroma of degraded images and other human vision-related factors.After analyzing the reasons for the degradation of foggy images,this essay is to study about the haze removalalgorithm based on the atmospheric scattering model from the perspectives of whether prior knowledge is required.Furthermore,the main factors that affect the restoration of foggy images is analyzed and optimized.The main work and innovation are divided into the following two parts:1)As for the phenomenon of texture details loss and blurry borders after the image haze removal using Dark Channel Prior algorithm,it is proposed to restructure and recombine the structure and texture of the haze image so as to avoid mutual interference caused by simultaneous processing of structure and texture.To avoid the occurrence of deep jump in depth change area of the structure layer after haze removal,a dark channel based on super pixel is proposed to enable the areas with the same depth of field the same transmittance.Then the structural similarity constraint of the non-local total variation regularization model is adopted to reduce the depth difference of the transmittance edge.In order to avoid the distortion of the sky area caused by the invalid information of the dark channel,a tolerance mechanism is introduced to identify the bright area and optimize the transmittance of the area.Lastly,to preserve the detailed texture of the scene,a mask indicating the texture area is established in the texture layer,in which the high-frequency information is saved.The experimental results show that the algorithm can retain the fine texture of the haze image,and the edge structure of the scene is clear after haze removal,and no distortion appears in the sky area.2)As for the phenomenon that the prior knowledge is invalid in some occasions,it is suggested to use the natural scene statistics to obtain the NSS characteristics related to the haze concentration and establish a haze concentration model without relying on the prior knowledge.Moreover,a mathematical relationship between transmittance and haze concentration is to be established to minimize the haze density.Aiming to reduce the calculation complexity of the transmittance model,the sensitivity between NSS features and transmittance as well as the redundancy between NSS features are to be analyzed.Finally,the three characteristics of saturation,brightness and hue of the dark channel with the strongest transmittance sensitivity and the lowest redundancy are obtained.Based on thecharacteristics above,a transmittance optimization model is established to achieve the purpose of image have removal.The experimental results show that the algorithm can improve the factorsof haze images related to human vision such as contrast,saturation and brightness.