基于DCT特征点的感知图像Hash函数  

Perceptual Image Hash Function Using DCT-Based Feature Points

作　　者：唐振军[1] 戴玉敏[1] 张显全[1] 张师超[1] 

机构地区：[1]广西师范大学计算机科学与信息工程学院,广西桂林541004

出　　处：《广西师范大学学报（自然科学版）》2012年第3期135-141,共7页Journal of Guangxi Normal University:Natural Science Edition

基　　金：国家863计划项目(2012AA011005);澳大利亚ARC国家基金资助项目(DP0985456);国家自然科学基金资助项目(61165009;61170131);广西自然科学基金创新研究团队项目(2012GXNSFGA060004);广西自然科学基金资助项目(2012GXNSFBA053166;2011GXNSFD018026;0832104);广西科学研究与技术开发计划项目(桂科攻10123005-8);广西师范大学博士科研启动基金资助项目

摘　　要：本文提出一种基于离散余弦变换(DCT)特征点的感知图像Hash函数算法。具体地说,先对输入图像预处理,生成规范化图像。在此基础上,将规范化图像分块并进行二维DCT处理,利用DCT交流系数构造特征点。最后计算特征点的重心,用特征点与重心的欧氏距离生成Hash。实验结果表明本文算法对正常数字处理稳健并具有良好的唯一性。接收机操作特性曲线对比发现,本文算法性能优于3种现有的算法。This paper proposes a perceptual image hash function using discrete cosine transform（DCT） based feature points.Specifically,the input image is first mapped to a normalized image by preprocessing.The normalized image is then divided into non-overlapping blocks.Next,two dimensional DCT is applied to each block,and AC coefficients of each block are used to form a feature point.Finally,gravity center of the feature points is calculated and the hash is obtained by computing the Euclidian distance between the gravity center and each feature point.Experiments show that the proposed algorithm is robust against normal digital operations and reaches good discrimination.Receiver operating characteristics（ROC） curve comparisons indicate that the proposed algorithm outperforms three well-known existing algorithms.

关 键 词：图像HASH HASH函数 图像检索 拷贝检测 接收机操作特性 

分 类 号：TP391.411[自动化与计算机技术—计算机应用技术]