Research Of Splicing Images Detection Algorithm Based On BDCT Coefficient

With the development of information technology, digital images are more and moreimportant in people’s life. At the same time, the advancement of powerful imagemodification software makes the manipulation of digital images much easier. Therefore,how to detect the authenticity and integrity of an image correctly and efficiently becomes avery important topic.It has been proved that some statistical measures of BDCT (Block Discrete CosineTransform) coefficients contain useful information for image splicing detection. In thisthesis, we have given a deep research and discussion on the discriminative power of imagesplicing detection based on the BDCT coefficients. After analyzing how to applying BDCTcoefficients on image splicing detection effectively through theory and experiments, wehave analyzed different features on the first order and second order of BDCT coefficientsand derived the discriminative power of BDCT coefficients. Based on the discriminativepower of the relevance of neighboring BDCT coefficient pair, a new discriminative featurerepresentation is proposed which applies the maximum mutual information algorithm onconditional probability transition matrix of BDCT coefficient to maximize thediscriminative power. On the other hand, the algorithm also reduces the calculationcomplexity and the dimension of the feature and improving the feature representation bypre-grouping the transition of specific BDCT component pair. The experimental resultsshow that the proposed algorithm achieves an accuracy of91.2%on grey image databaseof Columbia University by using Support Vector Machine (SVM) as the classifier.