HEVC Double Compression Detection with Non-Aligned GOP Structures Based on a Fusion Feature with Optical Flow and Prediction Units

Abstract

Video double compression detection is an important part of video forensics. Optical flow method can effectively reflect the periodic anomaly caused by Relocated I-frame (which is also called I-P frame) inside double compression video. In this paper, a fusion feature OFRB-NIPU is extracted to detect double compression video under high definition HEVC coding standard, far more with group of pictures (GOP) estimation. Firstly, optical flow residual of background area and the number of intra prediction units are extracted from each frame of input video respectively. A background subtraction algorithm is used to reduce the impact of moving objects. Finally, Support Vector Machine (SVM) classifier for double compression and Periodic Analysis (PA) method for GOP size estimation are adopted respectively. Experimental results have demonstrated better performance of the proposed method in comparison to the state-of-the-art methods.