Description
Digital forensics deals with the scientific reconstruction of digital traces for use in a court of law. Over the past two decades, this field has seen a significant increase in the use of machine learning. Given that more than 80% of internet traffic is video, digital video forensics is an area of growing research interest. Video encoding parameters contain a lot of forensic traces that are currently not being fully exploited. This is primarily due to the absence of a reliable distance metric for structured video data.
The objective of this Master’s thesis is to compare various methods of presenting video stream data to a distance metric. To this end, the student first learns to understand what video streams consist of. Then the student conducts own experiments with videos of different encoders and content to determine the stability of the results obtained with the metrics. Ideally, with these insights, the student is able to construct a classifier to distinguish authentic and manipulated videos. In the thesis, the student documents their approach, assesses the metrics with regard to their stability, and provides recommendations on what to pay attention to when working with these metrics.