Angle margin adaptive face recognition model training method

Abstract

The invention discloses an angle margin adaptive face recognition model training method, which comprises the following steps of performing pre-training based on a face recognition data set to obtain a pre-training model; inputting samples in the face recognition data set into a pre-training model to obtain corresponding feature vectors; respectively solving a corresponding intra-class variance for each class in the data set; obtaining a corresponding angle margin according to the intra-class variance corresponding to each class; combining the angle allowance of each category and the addition angle allowance loss function to form a category angle allowance adaptive addition angle allowance loss function; and supervising the face recognition model for training by using a category angle margin adaptive addition angle margin loss function to obtain a final face recognition model which can be used for face recognition. According to the invention, on the basis of the angle margin loss function, the category adaptive angle margin algorithm is introduced, so that the trained model has higher accuracy and generalization ability.

Description

technical field [0001] The invention relates to the technical field of face recognition, in particular to a face recognition model training method with self-adaptive angle margin. Background technique [0002] At present, face recognition has become a very important biometric technology for identity authentication, and is widely used in many fields such as finance, commerce, and security. Traditional face recognition methods generally use descriptor operators to extract features from face images, such as local binary pattern (LBP) operators. This type of method is simple to implement, but the discriminative power of the extracted features is low, and the accuracy of face recognition is low. At present, mainstream face recognition models all use methods based on deep learning. The core of the face recognition method based on deep learning is the design of the loss function. In recent years, the proposal and continuous optimization of the loss function based on cosine distan...

AI Technical Summary

Problems solved by technology

[0004] Purpose of the invention: In order to overcome the deficiencies in the prior art, the present invention provides a face recognition model training method with an adaptive angle margin, which can set adaptive angle margins for different categories in the face recognition data set. to solve the problem of extremely unbalanced distribution between categories

Images