Research On Graphics And Image Enhancement Processing Based On GPU

With the development of human society,we have to face with the growing amount of information and data.In order to take full advantage of the big data,developing an efficient algorithm is particularly important.Nowadays,many researchers have put forward some complicated mathematical models in their own field,and these models usually come down to solve the large-scale equations.GPU parallel computing can solve these problem well.With the development of parallel technology,GPU architecture is constantly updating,and parallel programming is also getting easier,which brings convenience to the relevant programmer,which also greatly promotes the development of related industries.This paper focuses on the theme of fast surface modeling for terrain,normal vector of terrain triangular mesh and image autoregressive interpolation implemented on GPU.The core of those three methods is a time-cost approach,however all those algorithms have low internal coupling and are suitable for parallel optimization.Therefore,to enhance their execution efficiency,this paper makes full use of the GPU with lots of CUDA optimization strategies.The main contents of this paper are as follows:(1)In order to solve the problem of large amount of terrain data,this paper gives a 3D terrain real-time rendering method based on CUDA-Open GL interoperability.We make a lot of work to optimize the program with GTX680.Compared to the single-threaded CPU program in Intel Corei7-930,the running speed based on GTX680 is up to 212.3,well meeting the needs of real-time rendering.(2)We provide a new method for the interpolation of normal vectors according to the edgedirection interpolation algorithm for images.In theory,this algorithm is suitable for all complex surfaces.And many CUDA optimization strategies are used to enhance the performance of the GPU-Parallel interpolation of normal vectors using the edge-direction for terrain triangular meshes with a TESLA K80 GPU.Compared with a single-threaded CPU counterpart(Intel Corei7-930)in a host computer,the running speed based on the TESLA K80 is up to 646.4 times faster.(3)Image interpolation based on autoregressive models has achieved significant improvement in both PSNR and subjective visual quality of the reconstructed image.Because of the high time cost,image autoregressive-based interpolation algorithms are rarely used in industry for actual production.In our proposed approach,we modified the training window of the pixel,which is suitable for GPU-parallel optimization.In addition,we added some mathematical optimizations to the matrix operations.To optimize our algorithm,we used various CUDA optimization strategies and based on TESLA K80.Experimental results show that,while maintaining a high PSNR and subjective visual quality.Our algorithm achieves a high speedup of 147.3 for a Lena image and 174.8 for a 720 p video,compared to the single-threaded C CPU code in Intel Corei7-920.