基于可信修复的数字图像全局优化自恢复研究

数字图像内容真实性和完整性保护问题备受关注。自恢复技术在图像中嵌入与自身内容相关的信息，利用提取的信息定位篡改并恢复图像，可实现图像内容保护。但该技术提出至今在恢复性能方面尚存在缺陷，许多学术问题有待深入研究。本项目将自恢复问题等价为介于无损信息隐藏与盲修复之间的一般性问题，建立全局质量优化的自恢复模型，包括篡改与未篡改区域恢复质量的整体优化，嵌入数据后与篡改恢复后图像质量的均衡优化。并在此基础上实现自恢复的方法创新：改变基于分块灰度信息的传统方法，发展基于局部区域显著性特征的参考数据生成方法，并研究在自恢复中引入基于辅助数据引导信息传播强度和方向的可信修复，有效提高全局恢复质量；发展具有稳健性的动态自恢复方法，利用载体图像合适的稳健特征作为参考数据的承载对象，并通过分层数据分享及同步嵌入机制实现恢复性能随非恶意信号处理程度动态变化的方法。项目成果对图像认证和多媒体信息安全具有重要意义。

Image authentication and integrity protection have attracted the significant interests among the academia and the industry. Image self-recovery technique embeds the information related with image content into the image itself, and the extracted information can be used to locate the tampered regions and recover the image content. However, up to now, with regard to recovery performance, there are still many problems and shortcomings needed to solve in this field. In this project, we define the self-recovery as the general problem between reversible data hiding and image inpainting, and then, establish the globally optimized model for self-recovery quality, which includes the quality optimization for tampered region, non-tampered region and the embedded image. Based on this theoretical model, we will carry on the technical innovations for self-recovery: new methods for the generation of reference data based on local saliency features, new methods for the trusted inpainting by the propagation guidance of the assisted data, and new methods for the dynamical self-recovery with robustness through secret sharing mechanism and synchronous embedding. The research results have the important significance for image authentication and multimedia security.