提出一种基于认证安全性的视觉Hash设计方案.通过分析视觉Hash认证中鲁棒性和认证集合的相互关系,指出了采用多分辨率构造视觉Hash的必要性.多分辨率方法可以平衡视觉Hash认证中的虚警、漏警问题.设计了一种基于树状结构分解的视觉Hash算法.在图像经过树状结构分解之后,对各个分辨率级别的图像块量化提取鲁棒特征,进行精确Hash计算组合生成认证Hash.实验结果表明,本方案对JPEG压缩、滤波、噪声等处理具有良好的鲁棒性,Hash误比特率低于0.15,而对图像的恶意篡改有较强的敏感性,可以用于图像的真实性认证.
A perceptual image hashing scheme based on security of authentication is proposed. The relationship between robustness and authentication set is analyzed, and the necessity of using multiresolution in perceptual image Hash scheme is proved. The false negative probability and false positive probability can be balanced in authentication with multiresolution. A novel perceptual image Hash algorithm is designed based on tree structure decomposition. Image blocks with~various resolutions are quantified to extract the robust features, and the precise Hash algorithm is used to generate the authentication Hash. Experimental results show that, with a Hash bit error rate less than 0.15, the proposed scheme is robust against JPEG compression, filtering, noise, etc. And it is highly sensitive to malicious tamper and can be used in multimedia content authentication.