Realistic Texture Generation and Artistic Enhancement for CG Images

Computer-generated (CG) images has many applications in both computer games and films, but one concern about the computer generated(CG) images is that even if the 3D model is perfect, if the texture is not that real, then humans can still tell that the CG image is fake. The goal of the paper is that we can help artists automatically generate the textures from real photographs, based on the raw CG image, and then use the textures to help enhance the beauty of the CG image. The procedure is as follows: firstly, we retrieve the real photos from the manually collected photo database, based on the similarity between the CG image and the photographs; secondly, cosegment the CG image and the selected similar photographs, so that the corresponding regions can be separated and matched in one single step; Thirdly, from different segments, I retrieved the textures using constraint texture synthesis, that each synthesis step is from the similar segment parts. The final step, is enhancing the computer generated image using the generated textures from the previous step, by doing texture transfer. The method for texture transfer is using image quilting. This project is inspired by [Johnson et al. 2011], and the advantage of the system is that it does not require the 3D model of the CG image, just a raw CG image is enough, and besides, the generated textures can not only be used in the image enhancement, but also different pipeline of computer animation, like texture binding for a model and so on.