Research Of Foggy Image Clearness Technique

Under fog weather conditions, the images captured in outdoor are usually degraded by scattering due to atmospheric particles such as haze and fog. Those foggy images have the characteristics of low contrast, color distortion, poor details, etc. Thus, in order to improve the quality of image, it is necessary to research the fog-degraded image clearness techniques based on the analysis of image degradation mechanism.Fog removal for degraded image is a practical and hot problem in computer vision and image processing, which has a wide applicate prospect. Dedicated to the atmospheric physics model, the thesis developed a series of research, which mainly includes three aspects: image restoration method based on physical model, image enhancement method based on non-physical model and foggy video processing, and the image restoration method based on physical model is its key research object.The main contributions and innovations of this thesis are as follows:(1) Current image defogging algorithms are easy to produce the Halo artifacts in the edge of the scene depth and have weak abilities to preserve or enhance details, therefore, this thesis proposes a novel defogging algorithm based on the local extrema to overcome these drawbacks. The proposed algorithm utilizes atmospheric scattering model to realize the fog removal. It first uses region segmentation method based on generalized Gauss model(GGM) to obtain candidate region for solution of atmospheric light. Then, it applies the local extrema method to figure out three pyramid levels to estimate atmospheric veil, and manipulates the tone and contrast of details at different scales through multi-scale tone manipulation algorithm. The results of comparison experiment against traditional methods demonstrate that the proposed algorithm can achieve more accurate restoration for the color and details, weaken the Halo artifacts and has high execution efficiency.(2) Because current image defogging algorithms can not obtain ideal processing effect under dense fog conditions, this thesis proposes a novel algorithm to improve visibility for single image degraded by different concentrations of fog. At first, the estimation of atmospheric light and white balance are implemented. Then, the proposed algorithm constructs the model based on sparse gradient prior to estimate atmospheric veil, and restore the foggy image by inverting the atmospheric scattering model. Finally, according to the scene depth constraint, we construct an effective non-local regularization model to optimize the restored result. A comparative study and quantitative evaluation is proposed which demonstrates that our algorithm can remove fog from degraded images effectively, and has characteristics of contrast enhancement, noise suppression, etc. Compared with the traditional defogging algorithms, our algorithm performs better when processing the image degraded by dense fog.(3) Solution of atmospheric light has been influenced by white objects or other factors in fog scene, failure estimation of atmospheric light may cause the phenomenon of image details loss and color distortion. In view of the above problem, under the atmospheric scattering model, this thesis proposes a powerful and practical algorithm based on image fusion to remove fog from a single input image. At first, the proposed algorithm uses histogram statistics method to obtain the candidate collection of atmospheric light. Then, it uses the weighted least squares(WLS) algorithm to estimate transmission map. Finally, it uses an image fusion technique based on the pyramid decomposition to obtain the clear image. In the fusion process, it will refer to three factors: saliency, saturation and exposure. Experimental results demonstrate that the proposed algorithm can weaken the impact of estimation error of atmospheric light, achieve more accurate and natural visibility restoration and has high robustness and execution efficiency.(4) In view of the fact that atmospheric scattering model is not applicable to all fog scene, this thesis also makes research on image enhancement technology based on non-physical model. Through analyzing the characteristics of Retinex method in the process of foggy image enhancement, a Retinex enhancement algorithm based on Markov random field(MRF) model is proposed to improve single foggy image visibility. In HSV color space, the proposed algorithm first constructs an edge-preserving Gaussian-MRF model to estimate illumination component. The reflection component is obtained by Retinex principle, whose color and brightness is regulated through CLAHE algorithm. Then, it construct an effective Huber-MRF model to solve the problem of noise amplification. Finally, after the color space conversion, the proposed algorithm realizes the foggy image enhancement. The experimental results demonstrate that our algorithm has the characteristics of detail preserving, color restoration, noise suppression, and can enhance foggy image visibility robustly.(5) Through the study of relationship between defogging technique for single image and video processing technique, considering the characteristics of dynamic background video and static background video, this thesis proposes a fast defogging algorithm for foggy video processing. At first, the estimation of atmospheric light is implemented. In order to maintain the smooth of restored video and eliminate the tone mutation phenomenon, the proposed algorithm imposes spatio-temporal constraints on solution of atmospheric light. After coarse estimation of transmission map, the frame difference method is used to detect motion region, and it imposes spatial constraints on the calculation of coarse transmission map through the weighted fusion method. Then it uses the linear guide filter to refine the coarse transmission map. Finally, it obtains clear video by the atmospheric scattering model. Experimental results demonstrate that the proposed algorithm can improve the foggy video visibility efficiently and rapidly, solve the problem of video flash and maintain consistency in two adjacent video frames.