图像合成技术发展迅猛，在丰富人们文化创作需求的同时也造成了虚假图像及谣言泛滥成灾。如今的合成技术基本能够欺骗人眼感官，但是在更为精细的场景结构及规律层面仍然存在较多瑕疵，很容易被有经验的专家或针对性的取证方法识破，且缺乏较为客观有效的定量化质量评估方法。本项目研究基于场景精细感知对抗的虚假图像合成理论与技术，以场景信息精细感知为理论支撑，提出保持真实场景中精细结构及规律的高质量自动化图像合成方法，包括图像拼接、操控及生成等，并基于图像取证技术设计有效的合成质量评估方法及通过对抗进一步提升合成质量。通过本项目的研究，希望提高虚假图像合成的质量，发挥图像取证技术在合成研究中的评判和促进作用，进一步推动图像合成与取证的学科交叉及融合发展。

Image synthesis technology is developing fast, satisfying people’s demands for creative activities but in the same time making fake images and rumors a huge problem. The synthesis technology can now easily bypass the check of human eyes, but they still have many artifacts especially in aspects of more fine-grained structures and physical laws of the natural scene, which make them relatively easy to be identified by skilled experts or some image forensics methods. Another problem is the shortage of effective and quantitative methods for the quality evaluation of synthesized images. This proposal aims to study the theory and methods of image synthesis that is based on fine-grained perception of scene information and adversary. The effective sensing of scene information is the theoretic foundation of this proposal, and we propose some high quality automatic image synthesis methods that can keep realistic fine structures and laws of the scene. These synthesis methods include image splicing, manipulation and generation. This proposal further studies image forensics methods as an effective evaluation method for image synthesis and uses forensics as an adversary to improve image synthesis quality. Through this proposal, we hope to improve the quality for image synthesis methods, to introduce image forensics for the evaluation and improvement of synthesis and finally promote the cross study of image synthesis and forensics fields.