RST Invariant Image Forgery Detection

Background/Objectives: Copy-Move Forgery Detection (CMFD) is a very prevalent approach used to detect copy and pasted portions of the same image. The copied portion is rotated, flipped or scaled. The detection method should be invariant to rotation, scaling and translation. Many CMFD methods came into existence; however, some methods fail to withstand attacks such as Contrast adjustment, Gaussian blur and JPEG Compression. Although the methods are able to resist the attacks, they are computationally complex. This paper proposes a Rotation, Scaling, Translation (RST) invariant image forgery detection. Methods: Local Binary Pattern (LBP) is applied on the low frequency content of Discrete Wavelet Transform (DWT) decomposed image for feature extraction. Findings: The proposed method is invariant to rotation, scaling and translation attacks on the pasted portions of the image and able to resist post-processing attacks and has low computational effort. It is evaluated qualitatively and quantitatively on a CASIA database. Morphological operations are performed to reduce the false alarms. The correct detection ratio is in the range of 80% to 99% and false detection ratio in the range of 7% to 30%. Applications: There is a great demand to detect the forgery, which aids in the digital forensic analysis, in legal document substantiation, and various other fields.