Denoising is the critical step in digital imaging. This paper improved an image removal algorithm based on subpixel anisotropic diffusion and GPU accelerate technique. The subpixel difference of an image is explored into the diffusion equation (DE). The DE is solved numerically in the subpixel domain and accelerated by using GPU processing. A GPU implementation of the algorithm is presented to alleviate the time costs. The noise is removed using the diffusion procedure. The experimental results show that the proposed algorithm achieved better performance with same denoising results with less time consumptions the GPU acceleration improved the algorithm more suitable for real time application.