For the dynamic digital X-ray imaging system, in order to solve the degradation of DR (Digital Radiography) image quality, NL-means algorithm was introduced for the DR image denoising by using the Anscombe transform. In order to solve its complex calculation and time-consuming problem, GPU (Graphic Processing Unit) was used for its high parallel computing and fast floating-point calculation abilities. During implementation, the original noisy DR image was mapped to the GPU's texture and each thread calculated one pixel. By using its multi-threads, the NL-means algorithm could be effectively accelerated in GPU. The results show that the NL-means denoising method can effectively restrain the noise of dynamic DR images, and the acceleration with GPU can speed up more than 2 orders of magnitude without the loss of resolution, which met the satisfaction of real-time denoising.