Face alignment method

Abstract

The invention discloses a face alignment method comprising the steps as follows: collecting a certain number of face images as training samples and predicting samples, enhancing the images, and for the training samples, calibrating face key points and storing the location information of the key points; learning the training samples calibrated in S1 through a random forest algorithm to get the feature mapping functions Phi<t> of the calibrated key points and to further get the local binary features of the calibrated key points; and combining the local binary features of the calibrated key points obtained in S2 into a global binary feature, getting a global linear regression model Wt based on the feature and through global linear regression learning, and thus locating the face key points of a sample to be tested.

Description

technical field [0001] The invention relates to the field of face recognition, in particular to a face alignment method. Background technique [0002] Face recognition technology is based on the facial features of a person. For the input face image or video stream, it is judged whether there is a human face. If there is a human face, the position, size and each main face of each face are further given. The location information of the organs, and based on this information, further extract the identity features contained in each face, and compare them with known faces to identify the identity of each face. The whole process generally includes steps such as face detection, face preprocessing, face alignment, and face recognition. Face recognition algorithms mainly include algorithms based on facial feature points, template-based recognition algorithms, and recognition algorithms based on the entire face image. [0003] Face alignment is to align the face images detected in th...

AI Technical Summary

Problems solved by technology

[0022] When locating key point coordinates on the basis of face images, the accuracy of key point positions depends on the accuracy of manually calibrated key points on the face and the number of samples. ASM (Active Shape Model) is an algorithm for locating key points. To a certain extent, it can effectively locate key points, but this method does not take into account the grayscale information of the target object inside the outline, so when dealing with objects with rich grayscale information on the surface of the image (such as human faces) or images that are greatly affected by light , it seems powerless

This method requires a large number of sample sets to normalize the face image when preprocessing the image, and uses the PCA (Principal Component Analysis) method for the training set data. The process is cumbersome, time-consuming, and the key points of the face are located. Not very accurate, relatively rough, not practical, too time-consuming, unable to meet real-time requirements

Images