Research On Color Image Dehazing Algorithm

With the influences of smoke, dust, haze, fog among the atmosphere, imaging systems will suffer from low contrast and color distortion of the image detection results. Making use of image processing technologies to restore high contrast and colorful images from degraded images has been a focus in image processing areas. These problems are all called of image dehazing and have great significances in defense, industries and consumer electronics markets. With the advances in technologies, color imaging systems are being widely used, the methods of color image dehazing become an important research direction in signal processing of these systems. For this subject, this paper starts the research from the conventional color image enhancement dehazing technology and the model-based color image dehazing technology, analyses the advantages and disadvantages of each method, and improvement measures are proposed to specific problems.Firstly, Retinex Algorithm is studied in this paper and applied to the color image enhancement dehazing, three sub-kinds of algorithms are simulated. By analyzing these simulation results, the advantages, disadvantages and scope of application of each algorithm are confirmed. Secondly, the dark channel statistics prior dehazing algorithm based on Atmospheric Scattering model is researched. The halo effects along the depth saltation areas of the dehazing results are deeply analyzed and derived in this paper, and a modified dark channel statistics prior dehazing algorithm to this problem is proposed. Using edge detection technologies, the new algorithm makes a pre-correction on the original transmission coefficients. The simulation results show that the halo effects are effectively reduced and the color oversaturation of processing results is avoided. Thirdly, the superpixels theory of SLIC segmentation is introduced in the model-based de-haze algorithm, and a novel dark channel calculation method is presented. Combining the conventional dark channel calculation method and the clustering information of superpixels segmentation, the phenomenon of pixels’ extension of dark channel is effectively restrained, then the whole quality of a dehazing image is improved. Finally, all of the algorithms above are comprehensively evaluated from the subjective visual perception and objective comparison of images parameters. In the research of color image dehazing, the improvement measures proposed by this paper are positive and effective, but there still are some deficiencies.