HEVC Double Compression Detection Based on SN-PUPM Feature

摘要

During the process of video forgery detection, double compression is a significant evidence. A novel scheme based on the Sequence of Number of Prediction Unit of its Prediction Mode (SN-PUPM) is proposed to conduct double compression detection on videos under HEVC standard, together with estimation on GOP structures. Number of PU with three kinds of prediction mode (INTRA, INTER and SKIP) is firstly extracted from each frame inside a given video sequence. Then the SN-PUPM is calculated by Absolute Difference Values from adjacent three frames in original extracted features and filtered with Twice Averaging Filter to reduce noises induced by the process. Then, an initiative Abnormal Value Classifier is trained with SVM to label I-P frames and have a final sequence for double compression detection and GOP analysis. Nineteen original YUV sequences are adopted for dataset in experiments. Results have demonstrated better performance in HEVC double compression than previous method adapted to HEVC.