基于感知哈希学习的图像认证方法研究

With the rapid development of Internet and multimedia, multimedia authentication become the research hotspots of multimedia information security nowadays. Perception hashing is used for multimedia content identification and authentication through perception digests based on the understanding of multimedia content, which provides a more convenient way for solving the management problems associated with multimedia authentication. With the rapid growth of the un-authenticated data under the network environment and the increasement of user requirement, current methods are difficult to fit in with actual demand in efficiency, adaptability and security. This project mainly focuses on the research for image authentication method based on perceptual hashing. Specifically, the proposal consists of three components: 1) We carry out research on how to efficiently extract robust features based on posterior attack model and multi-view data. 2) In the basis of the former component, we further focus on effectively making use of multiple features to obtain accurate hashing code according to semantic-quantizer collaborative learning. 3) For better tamper detection and localization accuracy, we will design the calibration-perception hashing and reference-perception hashing co-authentication algorithms. This study will expand the application of multiview hash learning theory for image authentication, and the corresponding research results will provide new clues for other types of multimedia data authentication.

互联网的发展使得多媒体的真实性、完整性认证成为亟待解决的问题。感知哈希在理解多媒体内容基础上，通过简短的感知摘要来完成多媒体内容的识别和认证，为解决与多媒体认证相关的管理问题提供了一种更为便捷的方式。随着网络环境下大规模非认证媒体数据的快速增长以及用户对认证应用需求的不断增加，这些方法在高效性、适应性及安全性等方面均难以适应实际需求。本项目拟针对基于感知哈希学习的图像认证方法进行研究。具体包括：1）研究基于后验攻击模型与多视角信息的鲁棒特征提取方法；2）基于提取的鲁棒特征研究语义-量化器联合协同学习的感知哈希编码算法以提高二值编码表达的准确性； 3）研究校准-感知哈希协同认证与索引-感知哈希协同认证算法以提高篡改检测与定位的鲁棒性。本研究将拓展多视角哈希学习理论在图像认证领域中的应用范围，课题研究成果将为其他多媒体数据类型认证提供新线索。

互联网的发展使得多媒体的真实性、完整性认证成为亟待解决的问题。感知哈希在理解多媒体内容基础上，通过简短的感知摘要来完成多媒体内容的识别和认证，为解决与多媒体认证相关的管理问题提供了一种更为便捷的方式。随着网络环境下大规模非认证媒体数据的快速增长以及用户对认证应用需求的不断增加，这些方法在高效性、适应性及安全性等方面均难以适应实际需求。本项目针对基于感知哈希学习的图像认证方法进行研究。具体包括：1）提出了基于后验攻击模型与多视角信息的鲁棒特征提取方法；2）提出了基于语义-量化器联合协同学习的感知哈希编码算法以提高二值编码表达的准确性；3）提出了基于图像语义的感知哈希协同认证算法以提高篡改检测的鲁棒性。本研究将拓展多视角哈希学习理论在图像认证领域中的应用范围，课题研究成果将为其他多媒体数据类型认证提供新线索。

内容获取失败，请点击重试