A Study on Forged Face Recognition based on Improved YOLOV8

In this study, for the forgery face recognition task, the YOLOv8 model is improved based on the YOLOv8 model, and the multiscale null attention (MSDA) and SPD-Conv modules are proposed to enhance the model's recognition performance under complex forgery features. By introducing the MSDA module, the model is able to capture forgery features at multiple scales, which significantly improves the recognition of subtle forgery traces. Meanwhile, the SPD-Conv module enhances the model's sensitivity to low-resolution and high-compression images by preserving high-resolution features. The experimental results show that the improved YOLOv8 improves in precision, recall and mean average precision (mAP), with a precision of 84.5%, a recall of 80.2% and a mAP of up to 89.3%. These results demonstrate the effectiveness of the improved model and provide strong support for the practical application of the fake face detection technique.