Image spatial diffusion on GPUs

Image spatial diffusion targets on blurring small discontinuities while sharpening distinct edges. We propose two GPU-based methods to efficient generate content-aware spatial diffused images. In the first method, we apply an enhanced bilateral filter on the input image. Our filter adopts a spatial weight kernel in the form of an inverted Gaussian. This inverted-Gaussian-spatial bilateral filter (IGSB) can remove small spots in large smooth areas more efficiently than Gaussian bilateral filter that are commonly used. In the second method, we present a fast approximation of mean-shift algorithm on GPUs. Considering the parallel nature of GPUs, the process of mean-shift kernel density estimation is modified and improved to achieve satisfied color image diffusion effects.