Total posture face identification method based on complete binary posture affinity scale invariant features

Abstract

Disclosed is a total posture face identification method based on complete binary posture affinity scale invariant features. By the method, identification of faces with total posture variations (from horizontal 90-degree left turning to horizontal 90-degree right turning, head raising and head lowering) can be achieved. Experiments analysis shows that affinity scale invariant features are invariant to faces turning in a plane and invariant to faces turning within 25 degrees in a plane, and accordingly the method acquires multiple face images of specific postures, extracts affinity scale invariant features of the face images, and combines the features to generate complete binary posture affinity scale invariant features so as to realize invariance to total posture face turning. The method needs no simulation training, so that the method is much better than methods based on models in terms of computation complexity. Meanwhile, for conditions with large posture variations, the identification performance is much better than traditional methods based on features. The method has certain application value and significance.

Description

technical field [0001] The invention relates to gesture face recognition technology, in particular to the research and realization of a full-pose face recognition method based on complete binary pose affine scale-invariant features. Background technique [0002] With the 911 terrorist attacks in the United States and the leakage of network CSDN user information, biometric recognition technology has attracted everyone's attention, and face biometric identification and authentication technology has always been a hot spot in the field of biometric recognition. However, in practical applications, face recognition is often affected by many factors. When the face posture changes, the expression changes, the external light changes, and the face is blocked (wearing a scarf) , sunglasses), etc., the performance of face recognition will drop a lot, which restricts the practical application of face recognition. Among them, the impact of face pose changes on face recognition is that th...

AI Technical Summary

Problems solved by technology

[0002] With the 911 terrorist attacks in the United States and the leakage of network CSDN user information, biometric recognition technology has attracted everyone's attention, and face biometric identification and authentication technology has always been a hot spot in the field of biometric recognition. However, in practical applications, face recognition is often affected by many factors. When the face posture changes, the expression changes, the external light changes, and the face is blocked (wearing a scarf) , sunglasses), etc., the performance of face recognition will drop a lot, which restricts the practical application of face recognition

Among them, the impact of face posture changes on face recognition is that the intra-class difference between two faces of the same person is greater than the inter-class difference between two faces of different people, thereby causing recognition errors.

[0003] Therefore, many researchers are committed to the research of pose face recognition methods. Pose face recognition methods are mainly divided into two categories, one is feature-based methods, mainly Eigenface based on principal component analysis proposed by Turk et al., Belhumeur, etc. Fisherface based on linear discriminant analysis proposed by people, but the recognition performance of these two methods will drop a lot when the attitude changes greatly, so they cannot meet the requirements of full attitude recognition. Wiskott et al. proposed an elastic beam graph matching algorithm. The method needs to locate some key semantic feature points in advance, such as eyes, nose, and mouth corners, and the positioning of these key semantic feature points is itself a research difficulty. Ahonen et al. introduced local binary pattern LBP into face recognition, but Experiments have shown that this method can achieve a better recognition rate only when the attitude change is less than 15 degrees

Another type of method is the model-based method, mainly including the active shape model ASM and the active appearance model AAM proposed by Cootes et al., the 3D deformation model proposed by Blanz and Vetter, and the coupling factor analysis model proposed by Prince et al. The main disadvantage is that the model needs multiple face images for training, so the computational complexity is very high. In addition, like the active shape model ASM and the active appearance model AAM, there are also key points manually marked.

Images