Research Of Algorithm For Haze Image Restoration Based On FPGA

The effectiveness of outdoor vision systems are greatly limited in foggy conditions for the video and images captured by imaging device degrades because of the atmosphere scattering. Real-time recovery for the fogging video and images has important practical significance. This thesis studied the foggy image degradation model based on analysis of fogging image degradation mechanism. An effective defogging algorithm with low complexity is proposed. The algorithm achieved real-time clarification for video and images combined with parallel processing approach and the "pipeline" structure of FPGA. The main contents are as follows:1) The degradation mechanism of foggy image and the atmospheric scattering model were in-depth analyzed. Advantages and disadvantages of two image de-hazing algorithms based on atmospheric scattering were expounded after discussed in detail. All of these work made full theoretical preparations for following study.2) For dark channel prior algorithm has many shortcomings such as existing halos when processing sky area, high complexity and high resource consumption when implemented on the FPGA. A fast dark channel prior defogging algorithm is proposed. First of all, the phenomenon of halo in sky area was solved and also the hardware resource consumption was reduced by setting threshold value to atmospheric light A. Then optimized solving methods of transmission, transformed the serial computing of the transmission and atmospheric light to parallel computing. Finally according to the circuit design characteristics of the FPGA, introduced guided filtering instead of soft matting which complexity is unrelated to the window size, thus improved the efficiency of algorithm.3) A set of image de-hazing system based on DE2-70 board as hardware platform was developed in this thesis. The hardware circuits of fast dark channel prior algorithm were designed after designing of overall architecture, then fulfilled the transplant of the algorithm’s FPGA platform. Some important part of the modules, such as the dark channel module, atmospheric light module and Dehaze module were simulated. Finally achieved real-time foggy image recovery.4) The results of the algorithm’s MATLAB simulation show, the fast dark channel prior algorithm proposed in this article processes quickly, and the recovery image has good clarification and high visibility. The system’s highest frequency can reach 190 MHz in the hardware simulation, it is far greater than 27MHz which is the video signal data rate of ITU-R.BT656, this is fully meet the real-time requirements.