Source Camera Device Identification from Videos

Abstract

Source camera identification is an important and challenging problem in digital image forensics. The clues of the device used to capture the digital media are very useful for Law Enforcement Agencies (LEAs), especially to help them collect more intelligence in digital forensics. In our work, we focus on identifying the source camera device based on digital videos using deep learning methods. In particular, we evaluate deep learning models with increasing levels of complexity for source camera identification and show that with such sophistication the scene-suppression techniques do not aid in model performance. In addition, we mention several common machine learning strategies that are counter-productive in achieving a high accuracy for camera identification. We conduct systematic experiments using 28 devices from the VISION data set and evaluate the model performance on various video scenarios—flat (i.e., homogeneous), indoor, and outdoor and evaluate the impact on classification accuracy when the videos are shared via social media platforms such as YouTube and WhatsApp. Unlike traditional PRNU-noise (Photo Response Non-Uniform)-based methods which require flat frames to estimate camera reference pattern noise, the proposed method has no such constraint and we achieve an accuracy of $$72.75 \pm 1.1 \%$$ 72.75 ± 1.1 % on the benchmark VISION data set. Furthermore, we also achieve state-of-the-art accuracy of $$71.75\%$$ 71.75 % on the QUFVD data set in identifying 20 camera devices. These two results are the best ever reported on the VISION and QUFVD data sets. Finally, we demonstrate the runtime efficiency of the proposed approach and its advantages to LEAs.

Source Camera Identification with a Robust Device Fingerprint: Evolution from Image-Based to Video-Based Approaches

“…As illustrated by the confusion matrix in Figure 4, misclassification between devices of the same brand occur in most cases. Across multiple camera devices, our method consistently demonstrates improved identification accuracy compared to the state-of-theart MISLnet [43] and MobileNet [52] methods. To the best of our knowledge, the current benchmark accuracy for ISCI stands at 71.75%, as established by the CNN proposed in [52].…”

“…Across multiple camera devices, our method consistently demonstrates improved identification accuracy compared to the state-of-theart MISLnet [43] and MobileNet [52] methods. To the best of our knowledge, the current benchmark accuracy for ISCI stands at 71.75%, as established by the CNN proposed in [52]. Significantly, as highlighted in Table 6, our system, leveraging the new device-specific fingerprint, surpasses the existing best performance in individual device identification, achieving an identification accuracy of 79.58% in the experiment involving 20 devices from the QUFVD.…”

“…The experimental evaluation conducted by the authors demonstrated promising results, with the PRNU-net achieving an identification accuracy of 85.7% across 10 different camera models. Bennabhaktula et al [52] explored the relationship between network complexity and performance in source camera identification by employing CNNs with increasing levels of sophistication. The aim was to investigate whether certain components, such as constrained convolutional layers and residual noise images, were essential for accurate identification.…”

“…This approach allows us to focus on the key frames while reducing the computational complexity associated with processing the entire video sequence. Moreover, previous studies [39,42,52] on video-based source identification have highlighted the effectiveness of utilizing I-frames for camera model identification. To accomplish this, we utilize the FFmpeg library in Python to extract I-frames from the videos.…”

“…In this sub-section, we aim to assess the strength and significance of our new fingerprint in the field of video-based source identification. To achieve this, we compare the performance of our proposed system with other existing methods [43,51,52] that have demonstrated benchmark results on the QUFVD [43] through two experiments: source camera model identification (SCMI) and individual source camera identification (ISCI). Firstly, in Table 5, we compare the results of the experiment involving 10 different camera models present in the QUFVD dataset to evaluate the performance of our system in SCMI.…”