DeepMultiFace Forgery Dataset (DMFFD): A Comprehensive Dataset for Multifaceted Forgery Detection
Description
DeepMultiFace Forgery Dataset (DMFFD) presents a unique compilation designed for comprehensive research in the domain of facial forgery detection and verification. The dataset comprises 213 original images acquired from istock, which is an online royalty free, international micro stock photography provider. The images are then resized to a standard resolution of 300x300 pixels to ensure consistency across the dataset. To create forgery instances, 200 reference images from the standard Defacto dataset are employed. This dataset carefully combines the selected source images with reference images for generating forgery samples. Exceptional cutting-edge AI technologies, specifically Remaker AI and FaceSwapper AI, are used to produce 204 forged images. In order to further enhance the dataset's richness various image augmentation techniques, include horizontal flipping, random adjustments to brightness, contrast, gamma, hue, saturation, value, and Gaussian blur are applied to the forged images. As a result, the initial set of 204 forged images undergoes augmentation, yielding a total of 1224 images. The MultiFaceForgery Dataset offers researchers and practitioners a robust platform for exploring advanced techniques in facial forgery detection, verification, and mitigation, which serves as a valuable resource for ensuring facial recognition security.