Detection of Transcoding from H.264/AVC to HEVC Based on CU and PU Partition Types

Zhenzhen Zhang, Changbo Liu, Zhaohong Li, Lifang Yu, Huanma YanSymmetry 2019 · DOI

TL;DR

This paper proposes a novel method to detect HEVC videos that have been transcoded from H.264/AVC. By analyzing the unique characteristics of Coding Unit (CU) and Prediction Unit (PU) partition types, the authors achieve high accuracy in identifying these forged videos, demonstrating robustness against common video manipulation attacks.

What Problem Does This Paper Address?

The proliferation of video content and user-friendly editing tools has led to an increase in video manipulation, including transcoding. Forgers may transcode lower-quality H.264/AVC videos into the more efficient HEVC format to disguise their origins and gain popularity or profit. Detecting this transcoding process is crucial for video forensics and ensuring the integrity of digital media. Existing methods often focus on re-encoding within the same codec or struggle with the complexity introduced by cross-codec transcoding. This paper aims to fill this gap by developing a robust method to identify HEVC videos that were originally encoded using H.264/AVC.

Methods

The core of the proposed method lies in analyzing the structural patterns within HEVC video streams, specifically the Coding Unit (CU) and Prediction Unit (PU) partition types. These partitioning decisions are made by the HEVC encoder based on the video content and bitrate, and they reflect the video's compression history.

  1. Feature Extraction: The method extracts CU and PU partition types from I-pictures and the first P-picture of each Group of Pictures (GOP) in both singly compressed HEVC videos and transcoded AVC/HEVC videos.
  2. Mean Frequency Calculation: For each type of partition (I-CU, P-CU, I-PU, P-PU), the mean frequencies across the video are calculated. These form four distinct feature sets.
  3. Feature Concatenation: These four feature sets are concatenated to create a comprehensive 38-dimensional feature vector that captures the distribution characteristics of CU and PU partitions.
  4. Classification: A Support Vector Machine (SVM) classifier is trained using these feature vectors to distinguish between genuine HEVC videos and those transcoded from AVC. The SVM is chosen for its effectiveness in handling high-dimensional, non-linear pattern recognition problems.

The novelty lies in the combined analysis of both CU and PU partition types, recognizing them as complementary features that together provide a more discriminative signature of the transcoding process.

Datasets

The paper utilizes a combination of standard YUV video sequences for training and testing.

  • CIF Video Set: 18 low-resolution videos (352x288 resolution).
  • 1080p Video Set: 11 high-resolution videos (1920x1080 resolution).

To increase dataset size, each video was segmented into non-overlapping clips of 100 frames. This resulted in 43 CIF video clips and 32 1080p video clips. The HEVC encoding was performed using HM software, and AVC encoding/decoding used JM software.

Key Insights and Ideas

  • CU and PU Partitions as Compression Fingerprints: The paper highlights that CU and PU partition types are not arbitrary but are deeply influenced by the video content and the encoder's rate-distortion optimization. This makes them valuable "fingerprints" of the compression history.
  • Complementary Nature of CU and PU: While CU partitions reflect the overall complexity of a Coding Tree Unit (CTU), PU partitions offer finer details about prediction modes. Analyzing both together provides a more robust detection mechanism than relying on either alone.
  • Transcoding's Irreversible Impact: The transcoding process, involving decompression and re-encoding, introduces irreversible quantization and reconstruction errors. These errors subtly alter the video content, leading to different CU/PU partitioning decisions in the transcoded HEVC video compared to a video originally encoded in HEVC.
  • Bitrate Influence on Partitioning: The study demonstrates that even when the final HEVC bitrate is the same, the intermediate AVC encoding and subsequent transcoding can lead to distinct CU/PU partition patterns. This is particularly evident when comparing singly compressed HEVC videos with AVC/HEVC transcoded ones.

Key Results

The proposed method demonstrates high performance in detecting AVC/HEVC transcoded videos:

  • High Accuracy: Classification accuracy consistently exceeded 0.9 across both CIF and 1080p video sets, with a minimum accuracy of 0.904 and a maximum of 0.997.
  • Robustness: The method showed strong robustness against:
  • Frame-Deletion: Accuracy remained above 0.97 even after deleting 30 frames from 1080p videos.
  • Shifted GOP Structure: The method performed well when the GOP structure in the HEVC re-encoding differed from the original AVC encoding, achieving accuracies above 0.98 in most cases.
  • Comparative Performance: The proposed method outperformed a recent state-of-the-art method (Reference [16]) in most scenarios, particularly in its ability to leverage both CU and PU features, leading to an average improvement of about 1.1% in accuracy.

Limitations and Future Work

  • Special Bitrate Cases: The paper notes that in some specific bitrate scenarios, the CU partition types of singly compressed HEVC videos might appear similar to AVC/HEVC transcoded videos, potentially leading to a decrease in classification accuracy. This suggests that feature redundancy can occur.
  • Future Work: The authors suggest exploring other valid features to further improve classification accuracy. Additionally, they aim to investigate methods for estimating the codec parameters used in the first compression stage, which would provide even deeper insights into the video's history.

Summary updated

Abstract

High Efficiency Video Coding (HEVC) is a worldwide popular video coding standard due to its high coding efficiency. To make profits, forgers prefer to transcode videos from previous standards such as H.264/AVC to HEVC. To deal with this issue, an efficient method is proposed to expose such transcoded HEVC videos based on coding unit (CU) and prediction unit (PU) partition types. CU and PU partitioning are two unique syntactic units of HEVC that can reflect a video’s compression history. In this paper, CU and PU partition types of I pictures and P pictures are firstly extracted. Then, their mean frequencies are calculated and concatenated as distinguishing features, which are further sent to a support vector machine (SVM) for classification. Experimental results show that the proposed method can identify transcoded HEVC videos with high accuracy and has strong robustness against frame-deletion and shifted Group of Pictures (GOP) structure attacks.
BibTeX
@inproceedings{zhang2019detection,
  author = {Zhenzhen Zhang and Changbo Liu and Zhaohong Li and Lifang Yu and Huanma Yan},
  title = {{Detection of Transcoding from H.264/AVC to HEVC Based on CU and PU Partition Types}},
  booktitle = {Symmetry},
  year = {2019},
  doi = {10.3390/sym11111343},
}

Loading related content...

Figure