Research On Dehazing Algorithm Of Light Field Image Based On Multi-Cues Fusion

At present,computer vision systems has been widely used in surveillance,navigation and other fields,and the image quality captured by it has a significant impact on many applications in computer vision,such as image classification,object detection and traffic monitoring.However,under hazy conditions,due to the absorption and scattering of light by particles suspended in the atmosphere,the outdoor images obtained by computer vision systems often exhibit low contrast and poor visibility,and the images appear color distortion,which will reduce the performance of the entire system,and have a great negative impact on human security.Therefore,it is of great significance to study effective image dehazing algorithms.However,most of the existing image dehazing algorithms are difficult to extract the depth information of the scene to calculate the scene transmission,and when there has bright white areas similar to the haze color in the scene,the accuracy of global atmospheric light estimation will be significantly reduced.Therefore,In this thesis,we focused on the difficulty of extracting depth information of hazy scenes and reducing the accuracy of global atmospheric light estimation in bright white areas of the hazy scenes.The main work of the thesis is listed as follows:(1)To solve the problem that it is difficult to calculate the initial transmission by extracting the depth of the hazy scene,we present an image dehazing algorithm which combines light field multi-cues with atmospheric scattering model in this thesis.In this method,we use the light field multi-cues fusion depth estimation method to extract the depth information of the hazy scene and calculate the initial transmission.Then,in order to reduce the influence of noise on the transmission value of the object edge areas in the transmission map,we use the weighted 1-norm context regularization to construct the objective function to optimize the initial transmission iteratively.The experimental results show that the transmission value of the object edge area in the transmission map extracted by this method is more accurate,and the restored image can better retain the structure information and color information,which obtain better dehazing results.(2)When optimizing the initial transmission of the scene with the guided filtering method,some detail textures in the guide image that do not reflect the depth change will be introduced into the transmission map,resulting in a large error in the filtered transmission,we design a light field image dehazing algorithm based on mutual structure joint filtering in this thesis.we use the method of mutual structure joint filtering with structural consistency to optimizing the initial transmission map.Experimental results show that there is no detail texture that cannot reflect the depth change in the optimized transmission map,and the high precision transmission can be obtained to achieve better image dehazing results.(3)Aiming at the problem of when there has bright white areas in the hazy scene,the estimated global atmospheric light value often has a large deviation.By analyzing the characteristics of the sky region in outdoor hazy images,we design a global atmospheric light estimation method that combines depth and texture information in this thesis.Experimental results show that our method can eliminate the influence of bright white region effectively,and the estimated global atmospheric light value is more reasonable and accurate.