A Face Feature Point Initialization Method Based on Face Orientation Classification

Abstract

The invention discloses a face feature point initialization method based on face orientation classification, which belongs to the fields of image processing and computer vision. The implementation process of this method is as follows: firstly, for each newly input face image, its HOG feature is extracted, and its orientation classification label is obtained through random forest decision tree voting, and the mean value of the corresponding training subset feature points is selected as the its initialization value. Since there is little difference in the position distribution trend of face feature points under the same orientation, but there are obvious changes in the distribution trend of feature points under different types of face orientations, the mean value of feature points with the same orientation as the input face image is used The method of initializing can reduce the impact of face orientation changes on feature point initialization, thereby improving the accuracy of face feature point initialization.

Description

technical field [0001] The invention belongs to the fields of image processing and computer vision, in particular to a face feature point initialization method based on face orientation classification. Background technique [0002] Face feature point detection is an automatic face calibration algorithm. Its research purpose is to obtain a set of pre-defined feature point location information that is sufficient to describe the shape and structure of the face by analyzing the face image. As a basic research in the field of computer vision, facial feature point detection has important research significance and application prospects. For many face image related algorithms, such as face recognition, expression recognition, gender recognition, age recognition, face animation, video compression, etc., face feature point detection is often a necessary link. [0003] The traditional facial feature point detection algorithms mainly include active shape model, active appearance model ...

AI Technical Summary

Problems solved by technology

resulting in higher complexity of the algorithm

It can be seen that the random initialization algorithm is simple and fast, but due to the large difference in the distribution of feature points of the training samples under different poses, the random initialization cannot guarantee the effectiveness of the initialization when the face pose changes.

However, the method based on statistical learning is very dependent on the selection of features, and the initialization efficiency of progressive regression is not high.

Images