Event
PhD Dissertation Defense: Ravi Garg
Wednesday, August 28, 2013
2:00 p.m.-4:00 p.m.
Room 2211, Kim Engineering Bldg.
Maria Hoo
301 405 3681
mch@umd.edu
ANNOUNCEMENT: PhD Dissertation Defense
Name: Ravi Garg
Committee:
Professor Min Wu, Chair/Advisor
Professor K. J. Ray Liu
Professor Gang Qu
Dr.Avinash L. Varna
Professor Lawrence C. Washington, Dean's Representative
Date/Time: Wednesday, 28 August 2013, 2PM - 4PM
Location: Room 2211, Kim Engineering Building
Title: TIME AND LOCATION FORENSICS FOR MULTIMEDIA
Abstract:
In the modern era, a huge amount of digital information is available in the form of audio, image, video, and other sensor recordings. These recordings may contain metadata describing important information such as the time and the location of recording. As the stored information can be easily modified using readily available digital editing softwares, determining the authenticity of a recording is of utmost importance, especially for such critical applications as involving the use of multimedia recordings for law enforcement, journalism, and national and business intelligence.
In this dissertation, we study novel environmental signatures induced by power networks, which are known as Electrical Network Frequency (ENF) signals and become embedded in multimedia data at the time of recording. ENF fluctuates slightly over time from its nominal value of 50 Hz/60 Hz. The major trend of fluctuations in the ENF remains consistent across the entire power grid including when measured at physically distant geographical locations. We investigate the use of ENF signals for a variety of applications such as estimation/verification of creation time and creation location of a recording, and develop a theoretical foundation to support ENF based forensic analysis.
In the first part of the dissertation, the presence of ENF signals in visual recordings conducted in electric powered lighting environments is demonstrated. The source of ENF signals in visual recordings is shown to be the invisible flickering of such indoor lighting sources as fluorescent and incandescent lamps. The techniques to extract ENF signals from such recordings are developed to demonstrate that a high correlation is observed between the ENF fluctuations obtained from indoor lighting and that from the power mains supply recorded at the same time. Applications of the ENF signal analysis to tampering detection of surveillance video recordings, and forensic binding of the audio and visual track of a video are also discussed.
In the following part, an analytical model is developed to gain an understanding of the behavior of ENF signals.It is demonstrated that ENF signals can be modeled using a time-varying autoregressive process. The performance of the proposed model is evaluated for a timestamp verification application. Based on this model, an improved algorithm for ENF matching between a reference signal and a query signal is provided, and it is shown that the proposed approach provides an improved matching performance as compared to the case when matching is performed directly on ENF signals. Another application of the proposed model in learning the power grid characteristics is also shown. These characteristics are learnt by using the modeling parameters as features to train a classifier to determine the creation location of a recording among candidate grid-regions.
The last part of the dissertation demonstrates that there exist differences between ENF signals recorded in the same grid-region at the same time. These differences can be extracted using a suitable filter mechanism, andfollow a relationship with the distance between different locations. Based on this observation, two localization protocols are developed to identify the location of a recording within the same grid-region using ENF signals captured at anchor locations. The localization accuracies of the proposed protocols are compared. Challenges in using the proposed technique to estimate the creation location of multimedia recordings within the same grid, along with efficient and resilient trilateration strategies in the presence of outliers and malicious anchors are also discussed.
