The Research On The Algorithms Of Image Haze Removal

The quality of images captured in bad weather condition such as hazy weather is easily influenced by the scatting of ambient particulates, which affects the surface colour of image, and lower the contrast as well as fuzz up the object features. Therefore, not only the image viewing is decreased with a deterioration visual effect, but also affects the performance of all kinds of instruments which depend on the optical imaging system.However,the existing image defog technologies still have many limitations in the practical application.With the development of computer vision and the extensive application in many fields, such as transportation,military,security, the research on image deblurring in hazy weather and the enhancement of image quality is necessary. The main research of the thesis focouses on the single image dehazing algorithm base on the atmospheric scattering model, aiming to improve the image quality, algorithm efficiency and the scene adaptability as well, the major contents of this thesis are as follows:In-depth analysis on the algorithm of the dark channel prior image dehazing based on the atmospheric scattering model was performed, considering that the atmospheric light estimation is susceptible to be influenced by the highlight source in the image, a block method which can solve the problem of the dark illumination of the reconstructed image is proposed to find the accuracy atmospheric light. Since the block effect generates due to the estimation of transmission through minimum value filtering, which will produce halo artifact around the sharp edges in the dehazed images, the guided filtering instead of soft matting algorithm in the original algorithm is used to refine transmission map, which can mitigate the block effect effectively and reduce the time complexity. To cope with the problem that the dark channel prior becomes invalid for the sky regions, and the original algorithm may produce noise and color distortion in the process of image restoration, this thesis proposes a rule by selecting the threshold value as a function of the histogram of the image to process and repairing the sky domain transmission which are wrongly estimated, which shows remarkabale improvements on different kinds of images.The experiment results demonstrate that: compared with the original algorithm, the proposed algorithm enhance the performance of illumination and naturalness of image as well as reduce the complexity effectively.Taking the importance of the transmission map to receive the dehazing image into consideration, this thesis removes fog from a single image through the Markov random field to take the transmittance map which is base on the atmospheric scattering model. Firstly,the proposed algorithm obtaines the transmission map by assingning labels with a dedicate Markov random field, and optimizes the map estimation process through the graph cut-based ?-expansion technique. Subsequently, the guided filter was used to remove the redundant information in transmission map. Finally, the restored image was obtained according to the atmospheric scattering model.The experiment results show that: compare with the classical algorithms, the proposed method is more effective and robust, which yields high-contrasted and vivid defogging images.