Research On Real-time Haze Removal Based On FPGA

With the economy development,the air pollution is more and more serious,and haze has affected more and more cities.Haze degrades the image quality of the outdoor imaging equipment.To achieve a more clearer image,we should make the image haze removal and therefore it will be of significant importance that we develop the haze removal algorithm which is reliable and runs fast.At present haze removal algorithm can be categorized into two types,image restoration dehaze method based on the physics model and image enhancement dehaze method based on the haze image itself.Image restoration dehaze method based on the physics model utilizes the degraded model to estimate the parameters of model,it performs image degradation inversion and reconstruct the image.Image enhancement dehaze method based on the haze image itself utilizes image enhancement algorithm to remove haze directly without utilizing prior information about haze.At present haze removal algorithm is always realized by software in PC and it is a time consuming process.But in some special applications such as autonomous vehicles and access to battlefield information,it should remove haze as soon as possible.To make haze removal method real-time,we propose an improved method and realize it in the FPGA.We introduce the haze image degradation module and haze removal method using dark channel prior.Images after haze removal by this method is of good quality but the realization of refining the transmission map with guide filter is complex.Therefore we propose an improved haze removal method based on dark channel prior which can be realized in the FPGA easily.The main work are as follows:Firstly,we introduce the image restoration dehaze method based on the physicsmodel and the image enhancement dehaze method based on the haze image itself and analyze their advantages and disadvantages.By comparison we choose the haze removal method using dark channel prior.Images after haze removal by this method is of good quality but it is a time consuming method because the guide filter which is used to refine the transmission map should be re-computed after each filtering.We consider the refined dark channel as the scene depth and utilize it to estimate the transmission map,then we don't need to refine the new transmission map.Results demonstrates the new method provide a better image quality with around 40%computation time of He's method.Secondly,we design the Camera Link input module,DVI output module based on CH7301 C and DDR3 SDRAM controller module based on the Xilinx memory interface solutions(MIG)in Verilog.The realization of these modules lay a foundation for the realization of the improved haze removal method based on dark channel prior.At last,we introduce how to design the computation framework of the improved haze removal method based on dark channel prior in FPGA and depict how to realize four key components which are minI(x)module,3×3 and 15×15 dark channel module,binary mask module and dark channel synthesis module.