Steganalysis of Inactive Voice-Over-IP Frames Based on Poker Test

Jie Liu, Hui Tian, Chin-Chen Chang, Tian Wang, Yonghong Chen, Yiqiao CaiSymmetry 2018 · OpenReview

TL;DR

This paper introduces a novel steganalysis method for detecting hidden messages within "inactive" (silent) frames of Voice over IP (VoIP) calls. By analyzing the statistical properties of encoding parameters using a "poker test" and employing a Support Vector Machine (SVM), the method effectively identifies steganography, outperforming traditional entropy-based approaches.

What Problem Does This Paper Address?

Steganography, the art of hiding secret messages within seemingly innocuous data, poses a significant threat to network security. While much research has focused on steganography in images and videos, Voice over IP (VoIP) communication presents a growing challenge. Specifically, this paper tackles the detection of steganography within "inactive" frames of VoIP streams. These frames, often representing silence or low-activity periods, are attractive targets for steganography due to their abundance and the potential for high data embedding capacity without significantly impacting perceived audio quality. Existing steganalysis methods often struggle with these inactive frames, creating a gap in robust VoIP security.

Methods

The core of this paper's contribution lies in its novel steganalysis approach, which leverages the poker test and a Support Vector Machine (SVM).

  1. Theoretical Analysis of Steganography's Impact: The authors first theoretically analyze how steganography affects the bit distribution of encoding parameters within inactive VoIP frames. They posit that steganographic processes tend to make the distribution of '0's and '1's more symmetric. This symmetry, in turn, alters the frequency of specific subsequences of these parameter bits.
  2. Poker Test for Feature Extraction: The poker test, a statistical method for assessing the randomness of a bit sequence, is adapted to quantify these changes. By treating sequences of encoding parameter bits as inputs to the poker test, the authors derive statistical features that are sensitive to the symmetry introduced by steganography. Specifically, they calculate the poker test statistic for various parameter bit sequences derived from inactive frames. The paper demonstrates that this statistic tends to decrease as the embedding rate increases.
  3. SVM-Based Classification: These poker test statistics are then used as features for a Support Vector Machine (SVM) classifier. The SVM is trained to distinguish between steganographically modified frames and original, clean frames. The authors employ an SVM with a Radial Basis Function (RBF) kernel for its effectiveness in handling complex classification tasks.

Datasets

The experiments were conducted using a large dataset of speech samples:

  • Source: 2200 speech samples, each 10 seconds long, collected from language-learning lessons.
  • Characteristics: Samples were 8000 Hz sampled and 16-bit quantized, encompassing Chinese male/female and English male/female speech.
  • Focus: The study specifically focused on the inactive frames within these samples.
  • Steganography Application: Secret messages were embedded into these inactive frames using steganographic techniques described in Section 2.2 of the paper, at various embedding rates (from 0.1 to 1.0).
  • Encoding Standard: The speech samples were encoded using the G.723.1 codec, with experiments conducted for both its 5.3 kbit/s and 6.3 kbit/s modes.
  • Training/Testing Split: Half of the collected samples were used for training the SVM classifier, and the other half were used for testing its performance.

Key Insights and Ideas

  • Symmetry as a Steganographic Fingerprint: The core insight is that steganography in inactive VoIP frames introduces a detectable symmetry in the distribution of encoding parameter bits. This is a crucial observation that forms the basis of their detection method.
  • Poker Test's Sensitivity: The poker test, traditionally used for randomness assessment, proves to be a surprisingly effective tool for quantifying the subtle statistical changes induced by steganography in these bit sequences.
  • Inactive Frames are Vulnerable: The paper highlights that inactive frames, often overlooked, are prime targets for steganography due to their prevalence and the potential for high embedding capacity without noticeable audio degradation.
  • SVM's Power for Complex Features: The success of the SVM classifier demonstrates its ability to learn from the complex statistical features derived from the poker test, effectively distinguishing between legitimate and hidden data.

Key Results

The proposed poker test-based SVM steganalysis method demonstrated superior performance compared to the traditional entropy test:

  • Higher Accuracy: The poker test-based method consistently achieved higher detection accuracy across various embedding rates and both G.723.1 modes (5.3 kbit/s and 6.3 kbit/s). For instance, the poker test achieved accuracy levels comparable to the entropy test at much lower embedding rates (e.g., 60% vs. 100%).
  • Lower False Positive and Negative Rates: The method also exhibited lower false positive rates (FPR) and false negative rates (FNR), indicating more reliable detection.
  • Performance Correlation: Detection accuracy increased with the embedding rate for both methods, as expected.
  • Mode Comparison: The 5.3 kbit/s mode showed slightly better performance with the proposed method compared to the 6.3 kbit/s mode.
  • ROC Curves: Receiver Operating Characteristic (ROC) curves further validated the superiority of the proposed method, showing a better trade-off between true positive rate and false positive rate.

Limitations and Future Work

  • Focus on G.723.1: The study is specifically focused on the G.723.1 codec. While this is a common VoIP codec, its findings might not directly generalize to all other VoIP codecs without further investigation.
  • "Abnormality" at Low Embedding Rates: The paper notes an "abnormality" in FPR and FNR at low embedding rates, where the performance seems to degrade slightly before improving. The authors suggest this might be due to the initial ordered state of parameter bits and their slight disordering when few parameters are modified. Further research could explore this phenomenon more deeply.
  • Future Work: The authors implicitly suggest that extending this approach to other VoIP codecs and exploring more sophisticated feature extraction techniques could be valuable future research directions. The paper also opens the door for developing more robust countermeasures against steganography in VoIP communication.

Summary updated

Abstract

This paper concentrates on the detection of steganography in inactive frames of low bit rate audio streams in Voice over Internet Protocol (VoIP) scenarios. Both theoretical and experimental analyses demonstrate that the distribution of 0 and 1 in encoding parameter bits becomes symmetric after a steganographic process. Moreover, this symmetry affects the frequency of each subsequence of parameter bits, and accordingly changes the poker test statistical features of encoding parameter bits. Employing the poker test statistics of each type of encoding parameter bits as detection features, we present a steganalysis method based on a support vector machine. We evaluate the proposed method with a large quantity of speech samples encoded by G.723.1 and compare it with the entropy test. The experimental results show that the proposed method is effective, and largely outperforms the entropy test in any cases.
BibTeX
@inproceedings{liu2018steganalysis,
  author = {Jie Liu and Hui Tian and Chin-Chen Chang and Tian Wang and Yonghong Chen and Yiqiao Cai},
  title = {{Steganalysis of Inactive Voice-Over-IP Frames Based on Poker Test}},
  booktitle = {Symmetry},
  year = {2018},
  url = {https://openreview.net/forum?id=ORBHufMMoe},
}
Figure