Face-Swap Recognition via Head Area Occlusion

The patent applies universality by designing a single face-swap recognition model that can identify synthetic faces generated by multiple different automatic face-swap algorithms simultaneously. Instead of creating separate specialized models for each algorithm, the universal model processes images from various face-swap sources using a unified architecture, thereby reducing device complexity while maintaining broad recognition capability across different algorithm types

The patent segments the face area into multiple regions (e.g., left eye region, right eye region, nose region, mouth region) and applies different occlusion strategies to each segment. This segmentation allows the model to focus on specific facial features that are commonly manipulated by face-swap algorithms, improving recognition accuracy without requiring complex overall model structures

The patent extracts and isolates specific facial feature regions (eye regions, nose, mouth) from the complete face image by applying occlusion masks to other areas. This extraction process removes irrelevant background and peripheral information that could interfere with detection, allowing the model to focus computational resources on the most discriminative facial features where face-swap manipulations typically occur

The patent applies local quality by treating different facial regions with different levels of analysis and occlusion. Critical regions such as the eye areas and mouth receive focused attention through selective occlusion of surrounding areas, while less critical regions are occluded to reduce noise. This localized approach enhances detection precision by optimizing the analysis depth for each facial region based on its relevance to face-swap detection