基于特征学习的图像内容篡改被动检测及恢复研究

In recent years, fake image has become a concern in Internet, military and judicial areas and so on. The important method to judge whether the image content has been tampered with is passive forgery detection method. The existing passive forgery detection methods own some drawbacks, such as low robustness, low time efficiency, depending on only one image attribute, inaccurately located forgery regions, which are the bump for applying the passive forgery detection methods in real world. In this project, for achieving the efficient, robust and accurate passive forgery detection method, the research effort is focused on studying of the image feature, the image local feature abased on the orthogonal polynomial transform, the adaptive feature matching based on the approximately nearest neighbor search, modeling of the pixel level semantic. Moreover, the project further explore how to restore the tampered image to what it was, the different types of forgery will be restored by establishing the relationship between the image local sensitive hashing with the changes of image content. In the face of more powerful image forging technology, the successful implementation of this project will further improve the accuracy of image authenticity determination, reinforce the existing image content security protection mechanism, and accelerate the digital forensics technology applied in the field of Internet, military, judicial areas and so on.

近年来，伪造图像已经成为网络、军事和司法取证等领域中备受关注的问题，图像内容篡改被动检测是判定图像内容是否真实的重要方法。现有被动检测方法存在鲁棒性不强、时间效率低、过于依赖单一图像属性、篡改区域定位不准确等问题，这限制了图像内容篡改被动检测被投入实际应用。本项目围绕图像特征学习、基于正交多项式变换图像局部特征描述、自适应近似最近邻特征匹配筛选、篡改检测中像素级语义建模等关键理论与方法，实现高效、鲁棒、定位准确的图像内容篡改被动检测。此外，项目还进一步研究图像内容篡改恢复问题，通过建立局部敏感哈希值与图像内容变化的关联机制，实现不同类型的图像内容篡改恢复。面对图像伪造技术的日臻成熟，本项目的成功实施将提升判定图像内容是否真实的准确性，加固现有图像内容安全保护机制，加速数字取证技术在网络、军事和司法等领域的实际应用。

图像内容篡改被动检测作为图像取证领域的一个重要分支，近年来受到国内外学者的广泛关注。现存的被动检测方法在一定程度上能够有效的判定图像内容的真实性，但存在鲁棒性不强、时间效率低、过于依赖单一图像属性、篡改区域定位不准确等问题。本项目围绕图像特征学习、基于正交多项式变换图像局部特征描述、自适应近似最近邻特征匹配筛选、篡改检测中像素级语义建模等关键理论与方法，实现高效、鲁棒、定位准确的图像内容篡改被动检测。此外，项目还进一步研究图像内容篡改恢复问题，通过建立局部敏感哈希值与图像内容变化的关联机制，实现不同类型的图像内容篡改恢复。通过本项目的研究有望提升判定图像内容是否真实的准确性，加固现有图像内容安全保护机制，加速数字取证技术在网络、军事和司法等领域的实际应用。