Advanced techniques for ENF-based time stamp verification of audio and videos

With the development of technology; intelligent systems have rapidly become a part of human life. Consequently, millions of data circulates around us through these systems. These data, can easily be manipulated with the help of several digital tools. Indeed, superficial data, i.e., that doesn't exist can be generated using artificial intelligence. The electric network frequency (ENF) based media (audio or video) forensics is one of the most popular techniques to detect modifications in the media or meta data. Time-of-recording (time-stamping) verification, geo-location estimateion and media synchronization are some of the commen ENF-based forensic applications. ENF is the frequency of the electric voltage generated in the electrical grid. Under certain circumstances, time-dependent changes of ENF (ENF signal) are integrated into digital media. ENF signal can be estimated through time or frequency based methods, including zero-crossing (ZC), multiple signal classification (MUSIC), estimation of signal parameters via rotational invariant techniques (ESPRIT) and short-time Fourier transform (STFT), which is STFT, a frequency-based method, is one of the most widely used one out of these methods. In the STFT-based method, there are factors that affect the accurcay of ENF signal, and accordingly, the performance of the ENF-based forensic analysis, positively or negatively. Among these factors, determining the STFT parameters is critical. In this thesis; firstly, it was investigated how the selection of different STFT parameters affects the performance of media time-stamping applications depending on the file length. As a result of experimental application, the STFT parameters that increase the performance of ENF-based forensics were determined. Secondly, through different Auto-Regressive (AR(q)) model parameters, white noise signals were obtained from the ENF signals, and similar experiments related to time of recording were conducted by using these signals. Finally; an enhanced STFT-based ENF estimation method was proposed, to estimate the ENF samples in the first and last parts of the query media, which are not possible to obtain by the traditional STFT technique.