Enhancement And Restoration Algorithms For Fog-degraded Images

Image enhancement and image restoration aim at highlighting some details in the original image for human vision or computer analyzing and processing. In foggy weather, the contrast of images is drastically degraded mainly due to the presence of considerable number of atmospheric particles such as dust, mist and fumes, which makes some applications such as video surveillance, target tracking very sensitive to weather conditions.Many scientists have researched the clearness technique in bad weather conditions from two aspects including image enhancement and image restoration to improve the system robustness. Some achievements have been made in the past years. But the problem is so complicated and difficult that the current algorithms cannot work it well. There is still much work to do on it, so it has a great significance to study how to enhance or restore the fog-degraded image more efficiently.For these reasons, this dissertation mainly focuses on the analysis of image degradation mechanisms and the modification of some current algorithms and techniques to tackle the problem more effectively. In summary, our meaningful and detailed research works of this thesis are listed as follows:(1) To overcome the halo artifacts in the region where the illumination changes rapidly, which the traditional center/surround-based Retinex enhancement algorithm often suffers from, a modified Multi-Scale Retinex color image enhancement algorithm is proposed. The modified algorithm is more effective than the MSR algorithm in improving the contrast and preserving the color of the original image. In the modified algorithm, an adaptive anisotropic Gaussian filtering method is introduced, and the orientation of the long axes is determined according to the gradient orientation in the position. The modified algorithm is given and applied in the fog-degraded images enhancement. Finally, the experimental result comparison with the previous methods testified the validity of the modified algorithm.(2) A fog-degraded image enhancement method based on human visual system (HVS) is proposed. The parameterized logarithmic image processing (PLIP) model is introduced. The proposed algorithm utilizes the human visual system to segment the intensity components of the fog-degraded image into Devries-Rose region, Weber region, low-contrast and saturation region sub-images. With the modified contrast limited adaptive histogram equalization (CLAHE), the sub-images can be enhanced and fused to scale. The flow chart for the human-vision-based contrast enhancement algorithm is given. The defog experiments have been done to illustrate that the method can enhancement the contrast and details of fog-degraded images efficiently.(3) A single image restoration method to remove haze is presented. Atmospheric degraded model is analyzed. Using multiple images taken from foggy scenes with different densities, the traditional method can remove weather effects and produce reasonably good results. But the method requires that the images are taken from exactly the same point of view. The input requirement makes the method impractical, particularly in real time applications. The Dark Channel Prior (DCP) and the image restoration algorithm using DCP from a single input image are introduced. We modified the algorithm to improve the precision of atmospheric light estimation. The sky region is separated from the degraded image and the value of atmospheric light is estimation by the average of the 10% brightest pixels in the sky region. If the original image don't contain the sky region or the sky region can't be separated accurately, the value of atmospheric light can be derived from the atmospheric degraded model by the transmission map. The procedure of the proposed algorithm is presented. Some experimental results of fog removal and the comparison between different algorithms testified the validity of the proposed algorithm.In a word, the clearness algorithms is studied and explored in this dissertation, the traditional methods is modified and improved by several means. The experimental results on different images show the efficiency of the proposed algorithms.