Research On Objective Evaluation Method Of Dehazed Image Quality Based On Synthetic Hazy Images

Image is one of the most important transmission media in our daily life,whose quality can greatly impact the accuracy of the message and the visual perception.Outdoor images are usually degraded when captured in hazy weather,which prevents many computer vision applications from working normally.Hence,image dehazing,especially for a single image,gains extensive research.Various dehazing algorithms have been successively developed.But the study of evaluation method is still insufficient because of the lack of reference image in evaluation.Nowadays subjective and non-reference evaluation methods are commonly used in the performance evaluation,but the full-reference evaluation method which is more reliable cannot be applied to the evaluation of dehazed images.In order to solve the above problem,this paper proposes an objective evaluation method for dehazed image quality based on synthetic hazy images.The synthetic hazy images can be generated by atmospheric scattering model,and then we can obtain the reference images.During the study,we find the close connection between the generation of synthetic hazy images and the depth information of a single image.So we propose a single image depth estimation method,and this method is applied to the full-reference objective evaluation of dehazed images.Specifically,the innovative work of this paper mainly includes:(1)Research on the estimation of image depth map.Aiming at the problem that the existing methods do not take into account the particularity of the sky area and the insufficient utilization of the geometric feature information of the image,a depth map estimation method weighted by geometric priori and edge information is proposed in this paper.Firstly,a method of sky segmentation based on threshold and edge detection is designed to separate the sky region in the image,then the depth map based on geometric priori is created by using the geometrical feature information of the image,and then we combine two depth map together,including the depth map based on geometric priori and the depth map generated from deep learning,then the sky segmentation result is incorporated into the depth map,at last we get the final depth map result.From the experimental results compared with other methods,both subjective and objective evaluation results show the superiority of the proposed method.We can get a more accurate depth map of an image and make a good foundation of the research on the full-reference evaluation of dehazed images by using this proposed method.(2)Research on the full-reference evaluation of dehazed images.Aiming at the problem of the lack of full-reference objective evaluation method of dehazed images,this paper proposes a full-reference objective evaluation method based on image dataset.Firstly,several distinct images are selected,and then the depth map of each image is obtained by using the proposed method of image depth map estimation,and then we generate the synthetic hazy images based on atmospheric scattering model.The image dataset used in the evaluation consists of the original haze-free images and synthetic hazy images,and then the existing dehazing algorithms are tested by using this dataset.The full-reference objective evaluation indexes are used to evaluate the images,and then we obtain the evaluation results.According to the experimental results,the full-reference objective evaluation method proposed in this paper can obtain an accurate results which consistent with subjective feelings.Compared with other non-reference evaluation methods,the results obtained by our method are more accurate and objective.