Fogery image splicing detection by abnormal prediction features

The paper proposes an algorithm to expose photographic manipulation. Splicing photographic merges two or more parts from different photos into one composite. A well-tampered one may not be perceptible by eyes. However, even a good forgery can leave some subtle traces caused by forgery. The proposal algorithm firstly segments photo into several parts under perceptual grouping criterion, minimizing the disassociation between parts and maximizing combination within part with normalized cut algorithm. Then conduct the mean and standard variance features of inharmonic points, as well as 14 Haralick features, then fed them to a support vector machine(SVM) classifier. The test experiments show that the proposal method is effective in exposing large-size splicing photographic.