Improved Poisson image editing and its implementation on GPU

This paper proposes an improved Poisson image editing algorithm. It can improve the quality of the resulting composite between different images significantly by three steps. At first, we extract object boundary based on global Poisson matting. Then, optimal boundary is computed by shortest-path algorithm between the object boundary and user specified boundary. Finally, seamless composite is achieved by solving the Poisson equation defined by the objective function and the optimal boundary. Moreover, to improve the performance of the algorithm, we implement it on GPU. Experimental results demonstrate the proposed method could generate seamless results even for images have complicated component. Also, it only needs a few interactions from user so that it is convenience to use. By mapping key steps to GPU, the performance of the method is improved considerably. With the increasing of image resolution, we can get more performance profit from our GPU-based implementation.