Research And Realization Of Gray Image Match Algorithm With CUDA

Image matching in digital photogrammetry, computer vision, remote sensing image processing, medical imaging, folders, and search a database of images, which are widely used. Today’s image processing technology is already quite mature, But how to improve the process of image matching speed and accuracy has always been a hot research field of image processing. In real life applications, real-time is very important for systems. Matching more fast becoming of great practical value, without influence of matching precision. In the case of mature algorithms, the algorithm itself is more difficult to improve. So rely on hardware, speed will be the implementation of the algorithm is very helpful.In the history of the development process of the GPU, its function has been increased more and more. NVIDIA and ATI has been at the forefront of technology.From the original drawing capabilities to now be able to process huge amounts of data in parallel, with the more powerful processing capability. Since the NVIDIA CUDA architecture is put forward, to now appear CUDA6.5, while supporting the C language, allowing developers to more easily programmed directly on the GPU.In this thesis, two algorithms of gray image matching are analyzed: NC algorithm(normalized cross-correlation matching) and SIFT algorithm. NC algorithm uses the gray value image to achieve image matching, the algorithm is simple in principle, higher matching accuracy. But because of the huge amount of the algorithm,resulting image match become slower match. So NC algorithm is suit for through hardware improve matching speed. As a matching algorithm based on feature points,SIFT algorithm can adapt to image scaling, illumination, rotation well. According to the algorithm of this paper, we propose parallel analysis and presented CUDA realization, given the corresponding test data and results analysis. Experiments show that algorithms’ implementation with CUDA can get better acceleration results.