Fog Degraded Image Restoration Methods Based On Physical Model

In recent years, foggy weather is becoming more serious than before. This will bring some problems to road video surveillance and traffic navigation systems. Therefore, image haze removal technique has important research significance. In this thesis, based on the research of atmospheric scattering physical model, we conduct research on the physics-based fog degraded image restoration methods. The main contents are as follows:(1) We firstly analyze the characteristics of foggy image, and conduct an in-depth study on the transmission and environmental light calculating methods. The two transmission calculating methods are based on independent component analysis and dark channel prior respectively, and the two environmental light calculating methods are based on the global highest intensity point and dark pixels ranking respectively. Analysis indicates that the traditional haze removal methods have some disadvantages such as they can’t handle dense haze image, the complexity is high and they can’t exclude the interference of white objects, etc.(2) In this thesis, we conduct an in-depth study on the method of single image haze removal based on guided filtering. Initially, this algorithm introduces a white balance operation to correct the image color deviation and get the environmental light indirectly, and then introduces an atmospheric light parameter which is optimized by the guided filtering to replace the calculation of the transmission. The complexity of the method is low and it has wider scope of application. However, this algorithm still has some drawbacks in the white balance operation and the atmospheric light optimization based on guided filtering. So, we will further optimize this method.(3) For the foggy images with serious color deviation, the traditional white balance method is disturbed by the scene’s color easily, this white balance effect is bad. Thus, in this thesis, we conduct the white balance operation on the foggy image by removing the environmental light chromaticity. From the perspective of the physical model, this method combines the inverse intensity chromaticity space with Hough transformation method to get the environmental light chromaticity, and then realizes the foggy image white balance in essence.(4) Since the single image haze removal method based on guided filtering just takes the smoothing factor as a standard to optimize the atmospheric light, the restoration effect exists some contradictions between depth mutation regions and texture regions. So, in this thesis, we propose a haze removal method based on local smoothing guided filtering, this method uses the scene depth information to construct an improved guided filter with local smoothing factor, so the optimized atmospheric light can reflect the variation of the scene depth accurately and thus for the images with discontinuous scene depth, our method can achieve better restoration effect.