high performance computing on graphics processing units: hgpu.org

Image denoising is a classical problem in image processing and has applications in areas ranging from photography to medical imaging. In this paper, we examine the denoising performance of an optimized spatially-varying Gaussian filter. The parameters of the Gaussian filter are tuned by optimizing a mean squared error estimate which is similar Stein’s Unbiased Risk Estimate (SURE), but is agnostic to the noise amplitude distribution. By optimizing the filter to local regions of the image, this technique preserves fine features such as edges while smoothing relatively flat regions of the image. The proposed method is evaluated through simulations on some standard images, and it gives an improvement in peak signal-to-noise ratio (PSNR) that is competitive to far more sophisticated methods. Even though the proposed method is computationally expensive, it exhibits considerable parallelism, which we exploit in a Graphics Processing Unit (GPU) implementation.