MULTI-POSE HUMAN FACE FEATURE POINT DETECTION METHOD BASED ON CASCADE REGRESSION

摘要

A multi-pose human face feature point detection method based on cascade regression. The method comprises: extracting a pose index feature and establishing the corresponding optimal weak regression device; performing corresponding initialization according to different human face pose orientations; using an SIFT feature of a human face image as an input feature of human face orientation estimation; obtaining the orientation of an input human face image according to a random forest human face orientation decision tree; using a feature point average value of a face-down training sample as an initial value of a feature point of the input human face image; and extracting the pose index feature of the human face image, inputting same into the optimal weak regression device, obtaining a distribution residual to update the current feature point distribution, and completing human face feature point detection. The method can achieve a stable human face feature point detection effect, and is suitable for various intelligent systems, such as a human face detection and recognition system, a human-computer interaction system, a facial expression recognition system, a driver fatigue detection system and a gaze tracking system.