Multi-angle face data acquisition method and device

Abstract

The invention discloses a multi-angle face data acquisition method. The multi-angle face data acquisition method comprises steps that an acquisition zone is entered through face identification, and a user is prompted to keep head up; a face image under a head-up state is identified, face characteristic points are extracted, a characteristic proportion value L1 is determined according to the face characteristic points, and the face data under a present attitude is stored; face images not under a head-up attitude are identified, a characteristic proportion value L2 is determined , the characteristic proportion value L1 and the characteristic proportion value L2 are compared, a face attitude is determined according to the relationship between the L2 and the L1, and the face data when the face attitude angle is in a preset section is stored; whether the stored face data in various angles covers all the preset angle sections is determined; if yes, the progress is over. The invention further discloses a multi-angle face data acquisition device. The method is advantaged in that the multi-angle face data acquisition progress is simplified, and integrity and accuracy of multi-angle face data acquisition are improved.

Description

Technical field [0001] The invention relates to the technical field of image processing, in particular to a method and device for collecting multi-angle face data. Background technique [0002] Face recognition technology is an important biological recognition technology that performs identity authentication by analyzing the displacement shape and position relationship of facial organs. The existing face recognition technology mainly relies on the extraction of facial features, and the extraction of facial features has a great relationship with the state of the face in different poses. The change of the pose can cause the difference in the scaling of the face image. This results in changes in the facial images of the same person in different poses. The complex changes of face pose can provide very rich three-dimensional information of the face, which makes the determination of face pose a key link and one of the important research contents in face applications. The current face...

AI Technical Summary

Problems solved by technology

However, the geometric model method uses the geometric structure to describe the face and establishes the corresponding relationship between the model and the face. The disadvantage of this type of method is that the accuracy is not high; the nonlinear regression method uses machine learning to establish a non-linear The linear mapping function is used to estimate the face pose. This method is greatly affected by face positioning errors and image quality changes; the feature space description method assumes that in some feature spaces, the face pose corresponds to some specific feature information. relationship, and the corresponding relationship can be found through training. The accuracy of the algorithm is related to the training samples and the feature space used to describe the face. The experimental results obtained are unstable.

Images