基于广义高斯分布的图像盲检测算法  

An Image Splicing Detection Approach based on Generalized Gaussian Distribution and Support Vector Machine

作　　者：孙韶杰[1] 吴琼[1] 李国辉[1] 

机构地区：[1]国防科学技术大学信息系统与管理学院,长沙410073

出　　处：《信号处理》2009年第9期1388-1392,共5页Journal of Signal Processing

摘　　要：图像拼接是图像篡改机制中简单并常用的方法之一。本文基于自然图像小波子带系数的统计分布符合广义高斯分布的假设,提出了一种拼接图像检测算法。提取小波细节子带系数对应的广义高斯分布模型参数以及模型预测误差为特征向量,采用支持向量机实现了对自然图像和拼接图像的有效分类。实验结果表明本文算法达到了平均88.76%的准确率,性能优于使用Hsu提出的基于相机响应函数的拼接图像检测算法。Image splicing is one of the simple and commonly used image tampering schemes. Based on the assumption that generalized Gaussian model is fit to describe the statistical distribution of wavelet details subbands of natural image, we present an image splicing detection approach. We extract the two generalized Gaussian model parameters and prediction error of each wavelet details subbands as feature vectors to discriminate natural images and spliced images effectively using support vector machine. Experimental results show the average accuracy rate to classify natural images and splicing images can achieve 88.76%. The detection performance of our method is better than that of the method using camera response function proposed by Hsu.

关 键 词：图像拼接 小波变换 广义高斯分布 支持向量机 盲取证 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]