Real-time rotation invariant human face key point detection method

Abstract

The invention discloses a real-time rotation invariant human face key point detection method. The method specifically comprises the following steps: performing random rotation cutting on face detection data, generating regression labels of positive and negative samples and a face frame, training a first-level network, detecting an original image in cooperation with an image pyramid, eliminating part of negative samples, and returning to the face frame; training a second-level network by using the left positive and negative samples, outputting a fourth classification of a human face orientation, further classifying and removing a part of negative samples, and performing human face correction according to a human face rotation category predicted by the network in a direction corresponding tothe human face rotation category; and adding five key points of a face, training a third-level network to further remove the remaining samples, and leaving an accurate face region frame and face keypoints thereof. The method has the advantages that the used model is small in size, high in speed, good in robustness and stability and particularly suitable for a mobile terminal, and the human facewhich rotates randomly in the plane can be efficiently detected.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a real-time rotation-invariant face key point detection method. Background technique [0002] The English name of face detection is Face Detection. The problem of face detection originally originated from face recognition (Face Recognition). The research on face recognition can be traced back to the 1960s and 1970s. After decades of tortuous development, it has become increasingly mature. Face detection is a key link in automatic face recognition system. Early research on face recognition mainly focused on face images with strong constraints (such as images without background), often assuming that the position of the face is always or easy to obtain, so the problem of face detection has not been taken seriously. [0003] With the development of e-commerce and other applications, face recognition has become the most potential biometric authentication method. This appli...

AI Technical Summary

Problems solved by technology

In the prior art, on the mobile side, it is not possible to effectively detect faces that rotate arbitrarily in the plane

Images