Developing a Feature Set from Scene and Texture Features for Detecting Neural Texture Videos Using Boosted Decision Trees

The prevalence of manipulated videos presents a significant challenge in today's era dominated by social media. Various types of fake videos, including notable examples such as Neural Textures, exist. Identifying such deceptive videos is a complex task. This research aims to understand the unique characteristics associated with Neural Texture videos. In the pursuit of comprehending these videos, the study explores the distinguishing traits that define them. The research employs techniques for scene and texture detection to formulate a distinct set of nineteen data features. This feature set is crafted to determine whether a video exhibits Neural Texture characteristics. To validate this set, a standard dataset of video attributes is utilized. These attributes undergo analysis using a machine learning classification model. The results of these experiments are assessed through four distinct methodologies. The evaluation reveals favourable performance outcomes when the machine learning approach and the proposed feature set are used. Based on these findings, it can be concluded that employing the suggested feature set enables the prediction of whether a video displays Neural Texture characteristics. This confirms the hypothesis that a correlation exists between a video's attributes and its authenticity, specifically in determining whether the video qualifies as a Neural Texture.