University of Twente Student Theses
Visualization of Deepfake detection using forensic methods
Kadolkar, K.C. (2025) Visualization of Deepfake detection using forensic methods.
|
PDF
4MB |
| Abstract: | Abstract—The increased availability and ease of use of deep learning frameworks have led to a rise in the generation and distribution of deepfake images and videos [11]. This has created a pressing need for algorithms capable of identifying and characterizing these manipulated media [12]. While deep learning frameworks are a popular choice for such tasks, their black-box nature makes it challenging to understand or explain the features leading to a classification decision [13]. This lack of explainability renders these methods unsuitable for scenarios where interpretability is essential [14]. This study explores two approaches for deepfake detection: forensic methods and deep learning models. The forensic methods focus on visualizing specific artifacts in manipulated media, such as inconsistencies in blending, color, and texture in targeted facial regions like cheeks and eyebrows. These techniques enhance in- terpretability by highlighting anomalies and facilitating an under- standing of manipulations through visual evidence. Conversely, the deep learning approach employs a Convolutional Neural Network (CNN) trained with K-Fold Cross-Validation on the cheek, eyebrow, and full-frame regions, prioritizing classification accuracy over interpretability. A comparative analysis reveals the strengths and limitations of each method, emphasizing the trade-offs between accuracy and explainability. The findings contribute to the development of hybrid strategies that combine the strengths of forensic visu- alization with the predictive power of deep learning, advancing the field of deepfake detection. This increases the explanability of deep learning models by not only looking at the larger picture but also at specific parts which contain artifacts that can then be highlighted using forensic method. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Computer Science MSc (60300) |
| Link to this item: | https://purl.utwente.nl/essays/105302 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page