Abstract
Poisson-based image editing technique has been popular in CG community. Natural intensity distribution of synthesized image is achieved by solving the Poisson equation with the modified gradient field copied form a source image. Non-photorealistic synthesized results may be obtained when a smooth source image is synthesized with a detailed or noisy target image. The conventional method incorporates multiresolutional random noise into the synthesized image, but it is not suitable for some images with geometric details. In this research, we propose a new framework for Poisson-based image synthesis by using the image analogy approach with the edge-preserving filter in oder to obtain more realistic results.
