Research On Image Dehazing Algorithm And Heterogeneous Parallel Implementation

During the smoggy weather,the decreasing of the visibility of the image will be severely reduced with the severity of the smog.Meanwhile,the smog and fog will lead the lost of the details and the insufficient of the contrast rate of the image.Thus,for video surveillance,traffic system,aircraft and spacecraft,it is important to improve the quality of the image during the smoggy weather.However,the algorithms that can remove the smog and fog,are limited by their processing speed,power cost and the development period when using them in reality.For these reason,a computing platform,which has the capability of parallel computing,based on the heterogenous system has been considered.The heterogenous computing platform has many advantages in several aspect.In this paper,our main work is to study the image-improving algorithm for the smoggy weather and its realization in the heterogenous-parallel method.First,studied the typical image-improving algorithm for the smoggy weather,including the Histogram Equilibrium and Retinex Theory,which are belonging to the image-improving method,and the dark-channel theory,which belongs to the image restoration method.Some simulation for the methods mentioned above and analyzed had been done,and their performance and parallelism had been analyzed.Then make some improvement for algorithm based on the darkchannel theory for its short come of the performance and the low computing quantity.Trough the comparison experiment,analyzed the improving-algorithm from the observation and data computation aspect compared with other algorithms.Secondly,studied the heterogenous-parallel computation.This part focused on the construction of two different computation platforms — — CPU-GPU platform and CPU-FPGA platform.Combined the Pascal frame of NIVIDA and GNC frame of AMD with CPU and analyze the performance of CPU based heterogenous parallel computation.As for FPGA based platform,analyzed the platform that combined with Intel-FPGA frame.Then,compared these three methods of computation.Next,studied the OpenCL model for programing.The research mainly focused on the four models of OpenCL using in the heterogenous-parallel computation.Finally,used the parallelized the dark-channel smog-removing algorithm and theoretically analyzed its computing quantity of each progress,in the ideal situation,with the serial algorithm.Then built a CPU-GPU computing platform,and programed the controlling program for the Host,and the OpenCL program to remove the smog,and studied the optimization method for the OpenCL program on this platform.Compared the processing speed and the quality of the result from the three platforms: CPU platform,CPU+GPU platform,and CPU-FPGA platform.Then,give an explanation of the difference between the results from those three platforms and give out our final conclusion.From the results,conclude that the performance loss of the image processed by the heterogenousparallel dark-channel algorithm is negligible;the CPU platform can improve the processing speed a lot,and the FPGA platform can somehow improve the processing speed,but not as significant as the CPU platform.