基于双树复小波变换统计特征的图像盲取证算法  

Blind Image Forensics Based on Dual-Tree Complex Wavelet Transform Statistical Features

作　　者：邓宇[1] 吴云洁[2] 周琳娜[3] 

机构地区：[1]北京航空航天大学自动化科学与电气工程学院,北京100191 [2]北京航空航天大学控制一体化技术国家级重点实验室,北京100191 [3]北京电子技术应用研究所,北京100091

出　　处：《系统仿真学报》2011年第8期1660-1663,共4页Journal of System Simulation

基　　金：国家自然科学基金(60970148)

摘　　要：双树复小波变换较常用的离散小波变换具有平移不变性和方向选择性的优点。用双树复小波变换对图像进行多尺度分解,分别计算分解后各层六个方向复高频子带系数幅值及幅值预测误差的一到五阶中心矩,并据此组合成对应的统计特征向量。利用该统计特征向量获取方法提取真实摄影图像和计算机生成图像的统计特征向量集,并采用支持向量机进行分类。实验结果表明,对测试集中的真实摄影图像和计算生成图像的识准率分别为98%和97%,取得了较好的分类效果。Comparing Dual Tree Complex Wavelet Transform with conventional Discrete Wavelet Transform supports that Dual Tree Complex Wavelet Transform has some advances including shift invariance and direction selection.Multi-scale decomposition of the image could be formed using Dual Tree Complex Wavelet Transform,and the first to fifth central moment of the amplitudes of wavelet coefficients and forecasting error was calculated individually based on six direction complex high-frequency subbands at each level of Multi-scale construction,then the statistical feature vector was composed of these moments.The set of statistical feature vectors of photorealism images and computer generation images could be extracted using the proposed feature vector extraction method,and then support vector machine could be used for the classification purpose.As result of the experiment,the corresponding accurate identification rate is 98% and 97% for photorealism images and computer generation images from the test sets individually.These results show that better classification performance of the proposed algorithm.

关 键 词：双树复小波变化 统计特征向量 真实摄影图像 计算机生成图像 图像盲取证 

分 类 号：TP309.2[自动化与计算机技术—计算机系统结构]