GPU-based Algorithm Of Haze Image Clearness And Parallel Implementation

As a research hotspot in the field of image processing,haze image clearness technology is widely used in video security,intelligent transportation,satellite remote sensing and other fields.Although the existing image clearness algorithm has achieved better defogging effect,there is still a bottleneck problem with high algorithm time complexity.A large number of complex calculations and high hardware requirements make it slow in engineering applications.For the haze clearness algorithm,the time complexity is high,and it is difficult to realize real-time processing.The paper proposes a method to improve the computing speed of the algorithm under the GPU platform,and uses the CPU+GPU collaborative processing method to improve the algorithm execution speed to meet the engineering application.First,the paper compares and analyzes the two mainstream general-purpose parallel computing platforms.One is the OpenCL general programming framework,and the other is the CUDA parallel computing architecture.The analysis focuses on the OpenCL architecture model used in the paper,as well as the memory model,platform model,programming model,and execution model.Secondly,the haze removal algorithm based on the principle of dark channel prior is the image clearing technology with better defogging effect at present,but it also has the problem of high time complexity of the algorithm,and the algorithm has defects in the acquisition of atmospheric light value.The paper analyzes all the steps of the algorithm and improves the way the atmospheric light value is obtained.The improved atmospheric light value is more robust than the original algorithm and is more suitable for parallel computing.Finally,the paper implements and optimizes the dark channel prior algorithm in parallel.In order to meet the huge computational requirements involved in the dark channel prior algorithm,the OpenCL programming framework is used to program the algorithm,and the algorithm is implemented in parallel on the GPU platform,so that the execution speed of the algorithm is significantly improved.Different memory allocation methods in the OpenCL model affect the access speed of the data.The paper compares different storage structures and selects the most appropriate memory allocation method to optimize the parallel program.In view of the fact that the GPU local memory has the characteristics of fast reading and writing speed,in the parallel implementation,the paper makes full use of the execution efficiency of the local memory optimization algorithm to further improve the processing speed.In addition,in the general programming of image processing,a large number of loop statements are usually involved.If the number of iterations of the loop is known in advance,the loop statement is expanded,which can effectively improve the execution efficiency of the algorithm.After parallel optimization,the algorithm execution speed is significantly improved,which lays a solid foundation for later engineering applications.