Research On The Implementation And Optimization Of The Motion Blurred Image Restoration Algorithm Based On GPU

Motion blurred images are very common in our daily life. As long as the object and the shooting system is in relative motion, there will appear this kind of phenomenon. So the deblurring technology is a very important step in image preprocessing.GPU has a huge memory bandwidth and powerful processing capacity for floating-point data. GPU computing has been used in many other fields. Comparing to CPU, GPU has accelerated the speed to one to two orders of magnitude in many fields. On the other hand, the CUDA architecture that the NVIDIA company launched lifted the shackles of the original purpose of the traditional GPU that bring to developers, and making the development of GPU programming easier, more powerful, and more extensive applications.In this paper, combining with the motion blurred image restoration algorithm and GPU technology. The first job in the paper is to implement the Fourier transform of two dimension images on CPU and GPU, these two different platforms. The second job of the paper is on CPU and GPU, achieving the Wiener filter for image restoration which the motion blur distance and motion blur angle are known. Then measured the GPU and CPU computing time, achieved a high speed-up. In order to increase these two algorithms’ degree of parallelization, using GPU to optimize the two algorithms from parallelization. This is the third job of this paper. In the case of the noise interference, the deblurring image after Wiener filter will be affected by the noise very much, and the forth job in the paper is proposing an improved Wiener filter algorithm. Under the circumstances of the motion blurred image affected by Gaussian noise or impulse noise, this improved Wiener filter algorithm can suppress noise effectively. Compared with the classic Wiener filter, the restoration images of this improved algorithm also have a better visual effect.