Loading...

Face Forgery Detection Through Statistical Analysis and Local Correlation Investigation

Asasi, Sobhan | 2023

90 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56788 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Ghaemmaghami, Shahrokh; Amini, Sajjad
  7. Abstract:
  8. Existing face forgery detection methods mainly focus on certain features of images, such as features related to image noise, local textures or frequency statistics of images for forgery detection. This makes the extracted representations and the final decision depend on the data in the database and makes it difficult to detect forgery with unknown manipulation methods. Solving this challenge, which is called the generalization challenge in artificial intelligence literature, has become the main goal of researchers in this field. In this thesis, the focus is on extracting effective features for success in forgery detection and preventing the performance of the forgery detection network from falling in the face of new data. The first goal has been achieved by extracting local and global features side by side. This step is done using the Conformer network which has two branches, transformer convolutional branches. Another goal of this thesis is answered by implementing an adversarial training method during the training of the proposed model. This adversarial training leads to the extraction of invariant features from real images. Due to the wide variety of forgery methods, only invariant features have been tried to be extracted from real face images. As a result of the training of the proposed model, both of the aforementioned goals are answered. In order to check the results and performance of the proposed model, two different comparisons have been made, Intra-testing and cross­-testing. In cross­-testing, the ability of the trained model has been evaluated against different falsification methods. The obtained results show that the proposed model is more successful than other models and modern networks in detecting forgery. Also, the investigations carried out in cross­-testing show that the proposed model has less performance loss than other modern methods in dealing with new forgery algorithms. The investigations obtained from the activation map of the proposed model indicate that some points of the face input images are of interest to the network and have been falsified and manipulated
  9. Keywords:
  10. Forgery Detection ; Generalization Challenge ; Adversarial Training ; Convolutional Neural Network ; Transformers ; Global Featur ; Local Feature

 Digital Object List

 Bookmark

...see more