Copy-Paste Forgery Image Blind Detection Algorithm Based on Histogram Invariant Moments

• Published in: Sensors & transducers Vol. 161; no. 12; p. 92
• Main Authors: Zhong, Junliu, Gan, Yanfen
• Format: Journal Article
• Language: English
• Published: Toronto IFSA Publishing, S.L 01.12.2013
• Subjects: Algorithms; Blinds; Dictionaries; Digital signatures; Efficiency; Forgery; Histograms; Image detection; Invariants; Mathematical analysis; Noise; Principal components analysis; Reproduction; Vectors (mathematics); Wavelet transforms
• ISSN: 2306-8515; 1726-5479; 1726-5479

Considered that general detection algorithms against common copy-paste image forgery have poor robustness, this article proposes a forgery image blind detection algorithm based on histogram invariant moments. The algorithm first carries out discrete wavelet transform on the image to be detected to extract low frequency part, then divides the low-frequency image into blocks. After that, histogram invariant moments characteristic vectors of these blocks are extracted, and a characteristic matrix is constructed using these vectors. The characteristic vectors in the matrix are sorted by dictionary, and finally the sorted adjacent blocks. The block is judged by confidence distance to determine whether some of them are copy-paste image blocks. Experiments show that, the proposed algorithm can more accurately locate the forged area of copy-paste images, and has better robustness on anti-noise, anti-compression, anti-rotation and anti-scaling. Meanwhile, the amount of computation is effectively reduced and detection efficiency is improved.