Deep Fake Lab

Abstract

The increasing prevalence of deep fake videos, which are digitally manipulated videos that falsely depict individuals saying or doing things they never did, has led to a need for reliable detection methods. In this project, we propose a deep fake detection technique based on hashing, which involves generating a unique fingerprint of a video frame or image that can be used to detect any modifications. Our approach involves extracting frames from a given video and applying perceptual hashing techniques to each frame to generate a hash. Perceptual hashing involves generating a unique signature that captures the perceptual characteristics of an image or video, such as its color, texture, and shape. We compared the hash of each frame to a pre-existing database of hashes for authentic videos, and calculated the similarity score between the two hashes using the Hamming distance. If the similarity score was below a certain threshold, the frame was considered a deep fake. To evaluate the performance of our approach, we used a dataset of both authentic and deep fake videos and calculated the detection accuracy and false positive rate. Our results demonstrate that our technique achieved high detection accuracy while keeping the false positive rate low. Our approach offers a simple and efficient solution for deep fake detection, which can be easily integrated into existing video analysis pipelines. Our method can be used to detect deep fakes in real-time, making it useful for applications such as video authentication and online content moderation. Overall, our project provides a valuable contribution to the ongoing efforts to combat the spread of deep fake videos, and we believe that our technique has the potential to be an effective tool for detecting deep fakes in various settings.