Image Dehazing Based on Haze Degree Classification
Xing Zhao, Ting Zhang, Wenxin Chen, Wei Wu · 2020
The existing technology of image dehazing neglects whether there is haze in an image and dehazes directly, which limits the efficiency of dehazing technology in real-world. Therefore, the classification of haze degree for image needs to be solved. This paper compares the features including brightness, haze factor, sharpness, etc. in clear and hazy images of different haze degree and obtain that different visual range images can be classified. And the Haze Image Classification Network is proposed which can classify an image to haze-free, haze-light or haze-dense. We test this method on our synthetic hazy dataset consisting of different visual range images and experiment results demonstrate the proposed idea performs well on accuracy, efficiency and real-time performance of hazy degree classification. We also train two dehazing model-haze-dense model and haze-light model for selecting and dehaze according to different haze degree. Compared to dehazing method without classification, doing haze degree classification before dehazing has got better performance on subjective and objective evaluation. The haze classification method and dehazing methods of this paper provide a good foundation for the following computer vision tasks, such as automatic driving, target detection, image recognition, etc.