Rapid Correction Method of Human Face Posture in Pitch Perspective

This paper proposes an innovative method for fast correction of the facial pose in the pitch angle under the limitation of single camera position in online communication, which causes perspective distortion of facial images. Based on facial prior knowledge, the proposed method identifies facial occlusions and designs an optimal key point selection scheme based on regular feature regions. Furthermore, the method achieves fast correction of foreground faces under new viewing angles by using the obtained paired key points, and realizes efficient background completion and edge blending based on sparse optical flow. This method effectively achieves the expected goal of correcting facial poses, relieving the non-real-time nature and realism gap of existing correction methods in online communication.