Copy-paste Forgery Detection Based on Non-negative Matrix Factorization

Copy-paste forgery is an important means of image tampering, an efficient and passive approach based on Non-negative matrix factorization is presented to detect the copy-paste forgery in the digital image. Firstly, image is segmented into overlapping blocks as the basic comparing unit, then DWT is applied to each sub-block and the low-frequency component is extracted. Non-negative matrix factorization of the low-frequency component is used to yield a reduced coefficient matrix which will be quantized subsequently as the features of sub-block. At last the detection is measured by the Jaccard similarity between each two sub-blocks. The experimental results show that the proposed algorithm is efficient and robust to the forgery detection of copy-paste image with less amount of computation and complexity.