图像感兴趣区域提取及加密研究

The openness of the network has brought a great challenge for image-data security transmission, and full encryption has been one of key technologies for such a problem. However, in many commercial multimedia applications like pay-per-view, they require part of multimedia content preview, where the regions of interest (ROI) are encrypted heavily and part of multimedia content is visualized or degraded to some poor quality so that it can be previewed free of charge to determine whether should pay for the corresponding service, which are beyond full encryption. In order to meet these kinds of multimedia applications, this project makes a research on ROI extraction with application to image encryption. We analyze and handle image feature data, and then integrate the important feature to construct some variational models for ROI extraction, and such regions can be detected by minimizing the energy functional. According to the physical characteristics of ROI, some adaptive encryption schemes are designed by the significant properties of optical transform and chaotic system, such as parallel processing, and sensitivity to control parameter and initial condition. The security and availability of image encryption scheme based on ROI will be verified by common attacks methods or provable security. This completion of this project will integrate image processing and image encryption and a new means of the future digital multimedia security applications will be achieved.

网络的开放性给图像数据的安全传输带来了很大挑战，图像完全加密技术一直以来是解决该问题的关键手段之一。然而，在很多商业多媒体应用中，需有意对用户感兴趣的图像区域重度加密，此时用户需要根据部分可视化区域来判断是否应该购买相应的服务，这种便捷的交互模式是完全加密技术力所不及的。本项目围绕这类特殊的应用需求研究图像感兴趣区域提取及加密技术。通过对图像特征数据的分析和处理，整合图像高效特征来构建面向感兴趣区域提取的变分模型，利用能量泛函的极小化来提取感兴趣区域；根据感兴趣区域的物理特点，充分利用光学变换并行处理的能力以及混沌系统初始敏感的特性来设计适配感兴趣区域的加密算法；同时，拟通过常用的攻击方法去验证或者利用可证明安全去证明感兴趣区域加密方案的安全性。本项目的研究实现了图像处理与图像加密的融合，其取得的研究成果将为未来数字多媒体安全应用提供了一种新的思路。

针对商业数字多媒体安全通信问题，本项目对图像感兴趣区域提取及加密技术进行了系统研究，主要研究内容与结果包括：（1）通过对图像特征数据的分析和处理，整合图像高效特征来构建了面向感兴趣区域提取的反应扩散方程；融合图像的局部和全局特征，提出了基于活动轮廓的图像边缘特征提取模型实现图像感兴趣区域提取。（2）根据图像感兴趣区域的物理特点，充分利用光学变换并行处理的能力以及混沌系统初始敏感的特性，设计了基于红外图像特征提取的选择加密方法；提出基于视觉意义加密图像的显著性区域保护方法；（3）利用常见的有潜力的攻击方式诸如已知明文攻击和选择明文攻击等方式去对现有的一些加密算法做了安全性分析工作。 本项目的研究实现了图像处理与图像加密的融合，其取得的研究成果将为未来数字多媒体安全应用提供了一种新的思路。