Research On Image Denoising And Its Parallelization On GPU

Image denoising plays an important role in image processing and computer vision image, its purpose is to remove noise to improve image quality in order to facilitate the subsequent processing or make the image have a good visual effect.Traditional denoising algorithm is simple in principle, and fast in operation, but the denoising effect is not obvious. Recently, a variety of excellent denoising algorithms have been proposed. And they have very big promotions in denoising effect. But these excellent algorithms are common computational complexity, need large amount of computation, which greatly limits the application of these algorithms.First of all, a detailed analysis of the dual domain filtering algorithm is made。 The dual domain filtering algorithm includes two parts, the base layer extraction and detail layer denoising. Aiming at the deficiencies of the original algorithm in base layer extraction,we studied the the WLS algorithm,and then we used it in base layer extraction. We made a better visual effect after image denoising.The base layer extraction and detail layer short time fourier transformation have very high data parallelism,but the algorithm itself is time consuming when sequential computing.So in this paper, we use open computing language OpenCL and the ability of data parallel computing in GPU to speed up the algorithm. The final result is that we made the algorithm 30 times faster than the original one, which greatly improves the practicability of the algorithm.