A framework for image inpainting is presented in this work. Inpainting is performed using discrete global optimization problem with an objective function rather than existing techniques. An iterative approach relies on extended belief propagation, called Priority BP has been used to solve the optimization. Although Priority-BP provides serious speedup by reducing complexity, inpainting an image is very intensive task. Hence, this study aims to provide further speedup improvements by using the power of GPUs through Compute Unified Device Architecture (CUDA). The effect of parallel processing framework is demonstrated on the real images for inpainting tasks.

Keywords : Image Inpainting, Image Restoration, GPU, CUDA.