Research On Single Image Dehazing Algorithm

Image is one of the main sources of information for people,and the quality of the image will directly affect people's analysis and judgment of information.In recent years,the frequent occurrence of haze weather in the North has caused low resolution and even serious distortion of images collected by imaging equipment,which has severely affected outdoor video surveillance and terrain surveys.Research on image dehazing is of great significance.This thesis firstly studies and analyzes the reasons for image degradation in haze weather,and conducts experiments,analysis and evaluation of various image dehazing algorithms,including single image dehazing algorithms based on image enhancement,image restoration and deep learning.Secondly,this thesis chooses to conduct an in-depth study on the single image dehazing method based on image restoration.The main results are as follows:1 The experimental analysis of the existing single image dehazing algorithm based on the dark channel prior is carried out.The results show that the current single image dehazing algorithm based on the dark channel prior has two main problems:First,the dark channel prior theory is not applicable to the sky area,the brightness of the sky area is relatively high,using the dark channel prior to the sky area will cause the sky part of the processing result to be distorted;secondly,although the use of guided filtering to refine the transmittance improves the computational efficiency,the processing result will produce a halo effect at the junction of the far and near scenes.2 Aiming at the problem of the distortion of the sky area in the processing results of the single image dehazing algorithm based on the dark channel prior,single-image dehazing via dark channel prior and adaptive threshold is summarized.The algorithm uses the dark channel-based adaptive threshold segmentation algorithm proposed in this thesis to effectively segment the sky part and the scene part of the image,and uses the segmented sky area to estimate the atmospheric light value.The guide filter algorithm is used to refine the global transmittance,and the adaptive compensation method proposed in this thesis is used to compensate the transmittance of the sky part to obtain the final global transmittance.The experimental results show that the thesis algorithm can effectively solve the sky distortion problem,and the experimental results have a significant improvement in objective aspects such as peak signal-to-noise ratio.This algorithm is suitable for situations where the image contains a large sky area and there is no sudden change in depth of field.3 Aiming at the problem of the halo effect in the processing results of the single image dehazing algorithm based on the dark channel prior in the sudden depth of field,an image dehazing algorithm via morphology and adaptive threshold is summarized.The algorithm first uses the sky area segmented by the adaptive threshold algorithm to estimate the atmospheric light value.Using gray-scale morphological opening and closing reconstruction operations instead of minimum filtering to denoise the RGB three-channel minimum map of the image to obtain the refined transmittance,and finally,adaptively compensate the transmittance of the sky to obtain the final transmittance.Experimental results show that the summarized algorithm has obvious advantages in operating efficiency,can effectively solve the halo problem,and improve the brightness of the sky.This algorithm is suitable for occasions where the image contains a small sky area,has a sudden depth of field,and requires high timeliness.