Research On SAR Image Enhancement Algorithm Based On CUDA

Synthetic Aperture Radar(SAR),a high-resolution imaging radar,is widely used in military and civil fields.Existing SAR image enhancement algorithms,which have high complexity,need long computing time while dealing with high-resolution images.In this thesis,Compute Unified Device Architecture(CUDA)based parallel algorithms of three basic functional modules in SAR image enhancement are studied and realized.An SAR image detail enhancement algoritlhm based on image stratification and noise visibility function,as well as the parallel implementation method based on CUD A,is proposed to solve the low real-time problem.The main work is shown as follows.(1)Meaning of SAR image enhancement is explained.Image degradation model and statistical distribution characteristics of SAR images are analyzed.Evaluation criteria of SAR image enhancement is given.Hardware architecture and programming model of CUDA used in this thesis are introduced.(2)Three functional modules in SAR image enhancement are studied.Parallel design and realization of existing SAR image enhancement algorithms are presented.Processing effect and acceleration performance of the parallel algorithm are analyzed.Results show CUDA based parallel implementation improve efficiency effectively.(3)Combined with the architecture of CUDA and characteristics of SAR image,an SAR image detail enhancement algorithm based on image stratification and noise visibility function is proposed.The algorithm uses bilateral filtering to stratify images.Gamma transform is used to adjust the contrast of the basic layer of image details and small fluctuations.Detail gain coefficient is controlled by noise visibility function.By this way,the noise is suppressed while the image details are enhanced.Results prove the proposed algorithm can significantly improve image in detail and in visual effect.(4)Feasibility analysis and parallelization design for detail enhancement algorithm based on CUDA implementation are received.Time consuming part is modified.Detailed steps of the algorithm are shown.Further optimization is done to improve parallel computing efficiency.Results confirm the algorithm based on CUDA can reduce computing time without decreasing the quality of image processing.Acceleration ratio reaches 75.756 for SAR image of 1024*1024.