GPU-accelerated SIFT-aided source identification of stabilized videos

Published 29 Jul 2022 in cs.CV, cs.MM, and eess.IV | (2207.14507v1)

Abstract: Video stabilization is an in-camera processing commonly applied by modern acquisition devices. While significantly improving the visual quality of the resulting videos, it has been shown that such operation typically hinders the forensic analysis of video signals. In fact, the correct identification of the acquisition source usually based on Photo Response non-Uniformity (PRNU) is subject to the estimation of the transformation applied to each frame in the stabilization phase. A number of techniques have been proposed for dealing with this problem, which however typically suffer from a high computational burden due to the grid search in the space of inversion parameters. Our work attempts to alleviate these shortcomings by exploiting the parallelization capabilities of Graphics Processing Units (GPUs), typically used for deep learning applications, in the framework of stabilised frames inversion. Moreover, we propose to exploit SIFT features {to estimate the camera momentum and} %to identify less stabilized temporal segments, thus enabling a more accurate identification analysis, and to efficiently initialize the frame-wise parameter search of consecutive frames. Experiments on a consolidated benchmark dataset confirm the effectiveness of the proposed approach in reducing the required computational time and improving the source identification accuracy. {The code is available at \url{https://github.com/AMontiB/GPU-PRNU-SIFT}}.

Abstract PDF Upgrade to Chat

Authors (5)

Citations (6)

View on Semantic Scholar

Summary

The paper introduces a GPU-accelerated SIFT method to invert video stabilization, enhancing the accuracy of source device identification.
It utilizes a two-phase approach with frame pre-selection using SIFT keypoints and grid search-based homography estimation to optimize efficiency.
Experimental results on benchmark datasets demonstrate reduced processing time and improved reliability, enabling real-time forensic analysis.

GPU-accelerated SIFT-aided Source Identification of Stabilized Videos

The paper "GPU-accelerated SIFT-aided source identification of stabilized videos" addresses a critical challenge in forensic video analysis, particularly the identification of the source device of stabilized videos. The key problem is the effect of video stabilization, commonly employed by modern devices, which complicates the forensic analysis by altering the intrinsic photo response non-uniformity (PRNU) patterns typically used for source identification.

Technical Overview

The authors propose a novel approach to solve the problem by leveraging the computational power of Graphics Processing Units (GPUs) alongside the Scale-Invariant Feature Transform (SIFT). The algorithm they developed inverts electronic image stabilization (EIS) by pre-selecting less stabilized frames and exploiting SIFT features to estimate camera momentum. This aids in identifying segments of the video that are less likely to have undergone substantial stabilization transformations, thus improving the accuracy of the identification process.

Methodology

The primary contributions of the paper lie in the optimization strategies employed for inverting the stabilization transformations applied to each frame. The authors propose a two-phase methodology:

Frame Pre-selection: The method begins by estimating camera momentum using SIFT keypoints, which helps in identifying the temporal segments that have undergone minimal stabilization. This step is crucial in reducing the search space for parameter estimation in subsequent steps.
SIFT-aided Frame Inversion: After pre-selecting frames, the algorithm performs a grid search to estimate the transformation parameters for each frame, using SIFT-based homography estimation to initialize these parameters efficiently. This approach ensures that the inversion process is both accurate and computationally efficient, thanks to the parallel processing capabilities of GPUs.

Results and Implications

Through experimental validation on a consolidated benchmark dataset, the proposed technique demonstrated its effectiveness in two main aspects: reducing computational time and improving source identification accuracy. The method's time efficiency, measured as Elaboration Time Per Second (ETPS), presents a significant improvement over existing solutions, thanks to the GPU's ability to handle parallel computations effectively.

The practical implications of this work are substantial. The improved accuracy and reduced computational cost enable real-time forensic analysis, which is critical in legal and investigative contexts where timely and reliable results are necessary. This method could be further refined by integrating deep learning techniques, which have shown promise in related tasks but require adaptation to handle the stabilization transformations inherent in video analysis.

Speculation and Future Directions

While the current results are promising, the paper hints at multiple avenues for future research. For instance, enhancing the model used to estimate camera momentum with more sophisticated keypoint selection strategies could further improve performance. Additionally, integrating neural network-based methods like Noiseprint could enhance the robustness and efficiency of the presented solution. Exploring fully video-based fingerprint extraction would also be a valuable advancement, moving beyond the current hybrid approach relying partially on still images.

Overall, the paper provides a substantial contribution to the field of video forensics through innovative use of GPU acceleration and SIFT features, setting a new benchmark for stabilized video source identification.

Markdown Report Issue