Shrinking the Semantic Gap: Spatial Pooling of Local Moment Invariants for Copy-Move Forgery Detection

Copy-move forgery is a manipulation of copying and pasting specific patches from to an image, with potentially illegal or unethical uses. Recent advances in the forensic methods for copy-move have shown increasing success detection accuracy robustness. However, images high self-similarity strong signal corruption, existing algorithms often exhibit inefficient processes unreliable results. This mainly due inherent semantic gap between low-level visual representation high-level concept. In this paper, we present very first study trying mitigate problem detection, spatial pooling local moment invariants midlevel image representation. Our method expands traditional works on two aspects: 1) introduce bag-of-visual-words model into field time, may meaning new perspective study; 2) propose word-to-phrase feature description matching pipeline, covering structure saliency information digital images. Extensive experimental results show superior performance our framework over state-of-the-art overcoming related problems caused by gap.