【作者】 韩琦；

【导师】 牛夏牧；

【作者基本信息】 哈尔滨工业大学 ， 信息安全， 2009， 博士

【摘要】 人脸图像是数字图像的一个重要子集,在社会生产生活的许多方面都有着广泛而重要的应用。然而数字图像处理技术的进步给人脸图像带来的诸多安全问题已经影响到公共安全等许多重要领域,这使得针对人脸图像的认证成为一个亟待解决的问题。人脸图像的安全问题主要在于人脸图像在应用过程中的真实性、完整性的保证以及传输过程中必要的保密性,传统的数据认证和加密手段很难满足其应用需求。人脸图像作为视觉信息的载体,其信息更多的以人类视觉感知系统所接受的感知信息的形式被传递、接收和评价,人脸图像认证本质上关注的是图像所传达的感知信息而非图像载体本身,因此如何定义、描述、提取人脸图像的感知信息,并且利用感知特征实现不同形式、不同粒度的人脸图像认证,是解决目前人脸图像安全问题的一种有效途径。本文从人类视觉感知系统的原理和特性出发,探讨了人脸图像感知特征的数字化表达方式及相应的鲁棒提取算法,进而分别从鲁棒Hash和数字水印两条技术路线,提出了一组能够实现不同方式和粒度的人脸图像认证及篡改感知内容恢复算法,为进一步增强人脸图像应用系统的安全性和灵活性,提出了一种基于感知内容的人脸图像分级加密方法,并将其应用在人脸图像安全传输系统。本文的主要研究工作和创新点在于:(1)通过对人类视觉感知系统中选择注意机制和感受野的层次、叠加特性等特性及其数学描述及模型的分析,提出了一组适合计算机处理的人脸图像感知特征表示方法,及两种感知特征提取算法:基于多尺度框架的感知点提取算法和基于Gabor函数的结构略图提取算法,分别实现了对人类视觉感知系统中心型感受野及其初级叠加的感受特性、视皮层简单细胞的感受特性的算法模拟,达到了较好的鲁棒性和感知内容描述性;(2)从数据集合映射的角度探讨了Hash概念从密码学到多媒体领域中的发展和变革,归纳了图像鲁棒Hash的典型模型。提出了三种人脸图像鲁棒Hash认证算法:基于结构略图的人脸图像鲁棒Hash认证算法、基于统计特征的人脸图像鲁棒Hash认证算法和基于感知点的人脸图像鲁棒Hash认证算法,针对不同的应用需求,实现了基于阈值匹配和精确匹配两种模式的人脸图像认证,以及像素块和扇形区域两种形式的篡改定位;(3)研究了基于数字水印的人脸图像感知内容认证技术,提出了利用鲁棒水印进行图像认证和利用多重水印实现语义层次篡改提示的人脸图像认证方法。提出了一种基于鲁棒水印的人脸图像认证技术,把基于Gabor特征的结构略图做为水印信息嵌入图像中,利用结构略图对内容变化的察觉能力和鲁棒水印对信息的隐藏和携带能力,构造了一个可以抵抗恶意篡改的鲁棒人脸图像认证方案;进而提出了一种结合鲁棒水印和半脆弱水印的多重水印认证算法,将反映人脸图像几何结构信息的特征点嵌入鲁棒水印,并利用半脆弱水印的定位能力发现篡改位置,最后结合这两方面的信息实现了语义层的人脸图像篡改提示;(4)研究了受篡改人脸图像的感知内容恢复技术与人脸图像感知内容分级加密技术,提出了一种基于整数小波分解的人脸图像篡改恢复算法,结合多重水印模型,实现了人脸图像内容的篡改定位与恢复;针对人脸图像分级加密问题,提出了一种针对索引图像的人脸图像分级加密算法,实现了依赖不同密钥的两级人脸图像加密,最后,根据分级加密算法的特点,结合公开密钥体制,提出了一种基于公开密钥体制的安全图像传输系统方案。更多还原

【Abstract】 Facial images are an important subset of digital images which have gained variousand critical applications in our society. However, many major fields such as public se-curity have been suffered threaten from the security issue of facial images brought bythe development of digital image processing techniques, which made the facial imageauthentication techniques turns to an urgently issues to be solved.The key issues of the security of facial image lie on the authenticity and integrityduring the use and the confidentiality during the transmission, the conventional methodsof authentication and encryption no longer satisfy these requirements. As a carrier ofvision information, the content of facial images is transmitted, received and evaluatedas the perceptual information which relays on the human vision perceptual system. Theimage authentication techniques should take more concern about the perceptual informa-tion transmitted by the image rather than the bit stream of the image file. So it would bean efficient approach to deal with the security issues of facial image that to define, de-scribe and extract the perceptual information of facial image, and to achieve facial imageauthentication with different pattern and granularity by use of the extracted perceptualfeatures.In this dissertation, the point of view that the facial image authentication techniquesshould focus on the”perceptual information”conveyed by the image rather than the car-rier itself is put forward by referring the principles and the characters of human visionperceptual system. Then, the expression and robust extraction methods of facial images’perceptual features are discussed. Further, two technical routers of facial image authenti-cation, the robust hashing way and the digital watermarking way, are investigated basedon the perceptual features of facial images. And the proposed algorithms achieve tamperrecovery based on the results of image authentication and tamper detection. To enhancethe security and ?exibility of facial image authentication system, a hierarchical encryp-tion algorithm is proposed.The major works of this dissertation is described as follow:Firstly, two most significant characters of human vision perceptual system, the se-lective attention machine and the hierarchical combination machine of receptive fields,and their mathematical descriptions or models are summarized and analyzed from thepoint of view of the constitute and the structure of human vision perceptual system. Then a set of perceptual features of facial image including their expressing methods andcorresponding extraction algorithms are derived. The multi-scale based perceptual pointsextraction algorithm simulates the center perceptual fields and their simple superimpos-ing. The Gabor function based structure skeleton simulates the simple cell’s perceptualfield.Secondly, the facial image authentication techniques based on robust hash are inves-tigated. The typical model of image robust hash is proposed. And three robust hash algo-rithms for facial image authentication are proposed, which include the structure skeletonbased algorithm, the perceptual point based algorithm and the statistical feature basedalgorithm. The threshold based matching method and accurate matching method for au-thentication are implemented in these algorithms. The block based tamper detection isachieved in the structure skeleton based algorithm and circle section based tamper detec-tion is realized in the feature point based algorithm.Thirdly, the digital watermarking based facial image authentication techniques areinvestigated. A novel robust digital watermarking based facial image authentication al-gorithm is proposed, which brings the robust digital watermarking into the field of imageauthentication. And then a multiple watermarking authenticating algorithm is proposedwhich embed the geometrical information into the robust watermarking and localize thetamper by embedding a semi-fragile watermarking in the interest region of the image– the face region. And a semantic report message about tamper could be drawn bycombining the two part information from the robust watermarking and the semi-fragilewatermarking.Fourthly, the recovering of tampered facial image and hierarchical encryption offacial image are investigated. To enhance the security and ?exibility of facial imageauthentication system, a tamper recovering algorithm and a hierarchical encryption al-gorithm are proposed. The tamper recovering algorithm based on the integer waveletdecomposition and the multiple watermarking model achieve the detection and recoveryof facial image. And the hierarchical encryption algorithm which aims to the palette im-age uses two keys to achieve partial and complete encryption. Further, the hierarchicalencryption algorithm is combined with a public key protocol to build a secure imagetransmission system.更多还原