A Face Verification Method Based on Difficult Sample Quadruple Dynamic Boundary Loss Function

Abstract

The invention discloses a face verification method based on a difficult sample quadruple dynamic boundary loss function, comprising the following steps: (1) constructing a face sample image database and preprocessing. (2) Group the face images in the database, and set an effective selection mechanism to select the four-tuple of difficult samples. (3) Set the model loss function as a quadruple loss function containing dynamic threshold boundaries, and input difficult samples for training to obtain a converged network model. (4) Use the trained network to extract the features of the face image to be verified, calculate the distance between the feature vectors, and judge the verification result according to the distance. The method of the present invention has the following advantages: 1) the model converges quickly; 2) the dependence on the manually set boundary threshold is low; 3) the face features are extracted more effectively and the accuracy of face verification is improved.

Description

technical field [0001] The invention belongs to the field of pattern recognition and relates to a face verification method, in particular to a face verification method based on a difficult sample quadruple dynamic boundary loss function. Background technique [0002] Face recognition technology is a biometric recognition technology based on human facial feature information. Face verification has always been a hot topic in the field of image recognition. It comprehensively uses machine learning, artificial intelligence, visual computing and other technologies, and is mainly used in access control and attendance systems, face verification security doors, public security, judicial and criminal investigations. [0003] Before the advent of deep learning technology, the classification method of face recognition mainly used hand-crafted features based on face image design. To distinguish them from deep neural network architectures, these method models are called "shallow" models....

AI Technical Summary

Problems solved by technology

This will lead to a large number of wasted samples during training, and the boundaries between classes are more difficult to distinguish

In addition, the distinction between easy samples (samples that are easily distinguished) and hard samples (samples that are difficult to distinguish) is also ignored, which obviously limits the expressiveness of the model and affects the efficiency of model training.

Finally, when setting the boundaries of positive and negative sample pairs, relying too much on manual experience will cause the convergence of the network model to be unstable.

Images