| Currently, most chinese regions have appeared a wide range of fog due to the wet weather and environmental pollution. The fog phenomenon has caused a great interference to the daily photos and reduces the definition of the captured picture. Due to the large data of the high resolution monitoring equipment, image defogging take long processing time and can not meet the engineering damand for real time. Therefore, image defogging is meaningful with high-speed parallel processing. In addition, the lack of objective evaluation criteria of image-defogging algorithms lead to the result that the defogging algorithms can't be evaluated precisely and guided easily.According to these problems, this paper mainly works on the following aspects:1) In this paper, we study the method and theory of image defogging, mainly focusing on the fog generating model, dark channel prior, softmatting algorithm. In the experiments, the color distortion and the massive effect of sky phonemon appear in the defogging image. In order to solve this problem, the method in this paper stretches the sky transmittance to modify the sky region and makes the enhanced image clear and lively.2) Facing HD video defogging in real time, the large data calculation consumes much time, and the algorthms and the processors both need to improve efficency. To solve this problem, we use multi-core processor platform(Tile-GX36) to defog image and designs the architecture and flow of parallel implementation. Moreover, we use the guided image filter instead of the softmatting algorithm to reduce time complexity. The Tile-GX36 processor platform inputs every image into the each processor, and 36 cores independently run the defogging algorithm. Experiment results show that the mutil-core processor platform greatly improve the processing speed. According to foggy video which frame images' size is 600*400, the frame rate can achieve 94.3fps and it is 17 times speed faster than the general PC. The result meets the engineering requirements for real-time.3) Evaluation of the image defogging algorithm focuses on the subjectivity and semi-reference image quality evalution methods and lacks of full reference image quality evalution. To solve this problem, we propose the method which use the depth information of reference images to add the artificial fog, and use defogging algorithms to enhance the fog image. Moreover,the full reference image quality evaluation index is proposed to evaluate the effect of the defogging algorithm in this paper. To construct the artificial fog image, firstly, the Fractional Order Darwinian Particle Swarm Optimization algorithm is proposed in this paper to divide the image into the remote shot, the medium shot and the close shot. Then different depth values are added to these shots, and the artificial fog is added to different parts of image using fog generation model. For the full reference image quality evaluation, we use four major classical algorithms to defog the artificial foggy image. Then, the full-reference image quality evaluation indexs(FSIM,SSIM) which can evaluate recovered results with the reference image is used in this paper. By contrast index values from each defogging algorithm,dark channel prior has the highest index value,expressed to the defogging image most similar to the reference image. Therefore, it is a good choice to use dark channel prior to defog the foggy image which can make the defogging image clear and lively.. |
PDF Full Text Request