Abstract:
Deepfake technology is a growing threat due to its strength to produce extremely realistic
manipulated videos, often used maliciously to disseminate misinformation or explicit
content. Destroying reputations and lives. Deepfakes are improving in terms of sophistication,
and developing reliable detection methods is now more crucial than ever. This
study provides a comprehensive literature review of current AI-based deepfake detection
techniques, examining their strengths, limitations, and real-world performance.
We reviewed existing methods’ challenges, especially in dealing with complex manipulations,
diverse datasets, and low-quality video data. Building upon this analysis,
the research focuses on enhancing the AltFreezing method, a respected approach for
detecting spatial and temporal artifacts in video forgeries. This approach leverages discrepancies
in motion and texture introduced during the deepfake generation process.
To improve its effectiveness, preprocessing t echniques a re introduced in the detection
pipeline. Methods such as noise reduction, contrast adjustment, and sharpening are
explored to refine t he i nput d ata, p otentially b oosting t he s ystem’s a bility t o detect
subtle manipulation artifacts. This integration was aimed at strengthening the performance
of AltFreezing, especially in challenging scenarios involving low-resolution content,
varying lighting conditions, or sophisticated forgeries that employ anti-detection
techniques. Preliminary findings indicate t hat i ncorporating preprocessing s teps may
enhance the overall robustness and accuracy of face forgery detection systems. These
insights contribute to ongoing research and development in the field of deepfake detection,
offering a promising direction f or f uture advancements in handling complex and
diverse real-world situations. The enhanced AltFreezing method. Preprocessing seems
to have the potential to become a vital tool for media platforms, law enforcement, and
cybersecurity professionals in the fight against malicious deepfake content.
