一种基于优化马尔科夫特征的图像篡改盲检测算法

文中主要针对拼接图像篡改检测,提出了一种基于优化马尔科夫特征的盲检测算法.该算法在传统马尔科夫特征的基础上,研究了不同相邻BDCT系数对的关联性对于拼接图像的检测能力,进而设计了一种基于互信息量最大化的加权BDCT系数转移概率特征;同时,通过对所有BDCT系数对进行预分组,降低了算法的计算量以及最终的特征维度.最后,采用支持向量机（SVM）作为分类器,在哥伦比亚大学提供的标准图像拼接库上完成测试,取得了较高的平均检测准确率（91.2％）,优于现有的代表性方法.

A Blind Detection of Splicing Images based on Optimized Markov Feature

An algorithm on blind detection of splicing images based on optimized Markov is proposed in this paper. By analyzing the BDCT coefficients, the proposed algorithm improves the traditional Markov feature by maximizing the discriminative power. A new discriminative feature representation applying the maximum mutual information algorithm on joint probability transition matrix of BDCT coefficient to maximize the discriminative power is proposed. On the other hand, the algorithm also reduces the calculation complexity and dimension of the new feature by pre-grouping the transition of specific BDCT component pair. The experimental results show that the proposed algorithm achieves an accuracy of 91.2% on gray image database of Columbia University by using Support Vector Machine （SVM） as the classifier.