Research On Fast Image Defogging And Implementation

Hazy weather is a common phenomenon, which appears frequently recently inour country. And the hazy weather always leads to the color distotion, low-contrastand the target detail disappearance in image processing, so the applicability of videomonitoring system is badly affected. At present, many chip manufacturers focustheir attention on the image haze removal. As a result, research on fast defogginghas strong engineering significance in practice. This dissertation improves an exitingimage defogging algorithm, designs a ASIC implementation architecture and teststhe improved algorithm on FPGA.This dissertation investigates the defogging method in detail based on the darkchannel prior theory. In view of the high time-temporal complexity and failure casesof this method, an improved method based on dark channel prior is put forward inthe dissertation. On the one hand, the bilateral filtering is proposed to refine darkcolor chart, instead of soft matting algorithm. The running time is reduced by15%compared with the original method. On the other hand, we use adaptive parametersto improve failure situation, and stability and applicability of the algorithm areenchanced. The experimental results show that the improved method can achievegood performance on defogging and reduce the failure situation.Based on the improved algorithm architecture, ASIC hardware implementationis developed and the RTL logic code writing and debugging is completed by VerilogHDL language. At the same time, a histogram statistics based on SRAM and TRAMimplementation architecture is presented. The architecture area is decreased by38%compared with the previous version, which can reduce costs obviously.In order to verify the correctness of the logic design, this dissertation sets upthe ASIC verification environment, analyzes RTL circuits, and gives simulationsequence diagram. Moreover, the online test on FPGA platform is completed. Testresults show that the designed method can well deal with haze images with goodperformance.