Research On Image Dehazing Algorithm And Its FPGA Implementation

In recent years,the development of computer vision technology has promoted the wide application of image processing technology in autonomous driving,smart city and other fields.In these applications,dehazing technology,as an important image enhancement technique,can improve the clarity and visual effect of the image containing fog,and make the images easier to analyze and process.The dehazing algorithm based on image restoration and deep learning has become the mainstream,while the algorithm based on deep learning requires a lot of computing resources and training time,which is difficult to meet the requirements of low-power processing.FPGA(Field Programmable Gate Array)has the advantages of parallel processing,low delay,low power consumption and strong real-time performance,which can reduce power consumption and cost while ensuring high performance.Therefore,this thesis studies the dehazing algorithm and its FPGA implementation.The main work is as follows:(1)The atmospheric scattering model and foggy imaging model are studied,and the key parameters affecting the defogging effect based on image restoration algorithm are obtained,which are atmospheric light value and transmittance.This thesis studies the dark channel prior dehazing algorithm,analyzes how to solve the atmospheric light value and transmission,and analyzes how the atmospheric light value and transmission affect the dehazing effect.(2)Aiming at the problems that the atmospheric light value is easily affected by the high-brightness light source,the calculation is large,and the accuracy of the transmission is low.,this thesis improves the solution of atmospheric light value based on dark channel prior dehazing algorithm,and proposes a transmission fusion algorithm to optimize the transmission solution.In order to improve the atmospheric light value,this thesis uses Gaussian filter to obtain the low-frequency components of the foggy image,uses the quartering method to obtain the region where the atmospheric light value is located,and solves the average value of the foggy original image in the region to obtain the exact atmospheric light value.The improved solution method of atmospheric light value solves the problem that the white highlighted object in the image with fog affects the estimation of atmospheric light value.The calculation of atmospheric light value is reduced.In order to improve the transmittance,anisotropic Gaussian filtering and Single-Scale Retinex were studied and analyzed to solve the transmittance.Through experimental analysis and comparison,the dark channel transmittance and Single-Scale Retinex transmittance were fused to obtain the fine transmittance.The improved transmittance solving method improved the de-fogging effect and improved the color effect of the image after de-fogging.(3)Based on the improved algorithm proposed in this thesis and combined with the FPGA hardware platform,the modularization design of the de-fog algorithm is carried out.RTL(Register Transfer Level)design and simulation verification were carried out for gray image solving module,image low-frequency component solving module,atmospheric light value solving module,dark channel image solving module,dark channel transmittance solving module,Single-Scale Retinex transmittance solving module and image restoration module.Finally,the algorithm proposed in this thesis was deployed on FPGA for implementation.And carry on resource consumption analysis,power consumption analysis.(4)In order to verify the effectiveness of the improved algorithm proposed in this thesis,MATLAB simulation experiments were conducted to compare with other defogging algorithms based on image restoration,and evaluation was conducted from subjective and objective dimensions;In order to verify the deployment effect of the proposed algorithm on FPGA,it is compared with several other commonly used application platforms,including de-fogging effect,power consumption and speed.Experiments show that the algorithm and actual deployment in this thesis have excellent performance.When de-fogging 640*480 images,it only takes 21 ms and the power consumption of the whole system is only 5.7W.The defogging effect is good and meets the engineering application requirements of fast speed,low power consumption and strong real-time performance.