Detecting Forgery Face Images Based on Local Feature Descriptors (SIFT)

Digital facial images represent people's identity and play a pivotal role in documenting events and their visual communication. Fake faces are detected using image processing and feature extraction techniques. The proposed methodology relies on a series of pre-steps that enhance the quality of visual representation and reduce computational complexity. The image size is reduced by half at each stage, the image is converted to grayscale, and the features are extracted using the SFT algorithm. A pre-trained deep learning model, EfficientNet-B0, was also applied to classify face images into real and fake ones. The FaceForensics++ database was used. The model was tested using noise added to images to evaluate its robustness and accuracy under different noise conditions. It was tested on salt-and-pepper noise and Gaussian noise, and horizontal geometric displacement, a type of image noise, was applied. Satisfactory results were achieved, with the accuracy of the proposed technique reaching 99.85%