Digital image falsification blind detection method based on third-order statistical features and combined classifier

The invention relates to a digital image falsification blind detection method based on third-order statistical features and a combined classifier. Firstly, true images and falsified images form a training set, a third-order statistical feature-condition symbiosis probability matrix is used for modeling intra-block correlation and inter-block correlation of a partitioning discrete cosine transformation coefficient of each image, feature data used for image falsification detection are extracted, and the feature dimension of the feature data is calculated according to 4 (2T+1) 3; secondly, combined classifier training is conducted, and basic classifier module files are stored; thirdly, feature data of tested images are obtained through the above method; lastly, whether digital images are falsified or not is detected through a stored basic classifier. The third-order statistical features are used for description of image content features, image falsification is detected based on the combined classifier, and therefore high image falsification detection correctness can be obtained, the digital image falsification blind detection method has a remarkable advantage in real-time performance compared with an image falsification detection method based on a support vector machine, and the practicability of digital image forensics is greatly improved.