基于自然图像统计特性的拼接图像检测算法

An Image Splicing Detection Algorithm based on Natural Image Statistics Characteristics

图像拼接被认为是最基本和最主要的图像非法编辑操作。拼接图像检测的关键问题之一是提取拼接图像有别于自然图像的区分性特征,将其转化为模式识别问题。本文从自然图像DCT变换系数的统计特性出发,分别应用高斯分布和广义高斯分布来建立其直流分量和交流分量的统计分布模型,同时结合图像小波变换系数的能量分布特性,提取模型参数和小波域的能量分布特性形成特征向量,送入支持向量机,实现对拼接图像和自然图像的分类和检测。实验结果表明,本文算法达到了平均80%的准确率,性能优于Ng提出的基于双相干特征的拼接图像检测算法。

In all illegal image editing operations, image splicing is considered the most fundamental and most important operation. One of the key issues of image splicing detection is to extract distinct features of spliced images, which are different from the nature of natural images, and then formulate it as a pattern recognition problem. In this paper, Gaussian distribution and generalized Gaussian distribution are applied to statistically model the DC （Direct Current） and AC （Alternative Current） DCT coefficients of natural images, combining energy distribution characteristics of wavelet coefficients of images. We extract the model parameters and energy distribution characteristics as feature vector. Then the feature vector is fed into the Support Vector Machine to classify natural images and spliced images. Experimental results show the average accuracy rate can achieve 80%. The detection performance of our method is better than that of the method using bicoherence features proposed by Ng.