Multi-Angle Face Detection Based on Improved RFCN Algorithm Using Multi-Scale Training

The R-FCN algorithm is mainly applied to general target detection, and is not a dedicated algorithm for face detection. In particular, most faces in real life have the characteristics of occlusion, blur, and changeable posture. The performance of the R-FCN algorithm needs to be improved. Therefore, based on the R-FCN algorithm, this paper proposes an improved R-FCN-FACE model to meet higher requirements for face detection. In the RFCN-FACE algorithm, the performance and detection effect of the face detection model are improved by introducing multi-scale training and online hard sample mining technology. A comparative test analysis is conducted on the benchmark data set Wider Face and FDDB. The experimental results show that the improved R-FCN-FACE model can show higher accuracy in multiangle face detection and improve the performance of the detection model.