Face recognition and attribute classification method based on multi-task convolutional neural network

Abstract

The invention relates to computer vision technology, which discloses a face recognition and attribute classification method based on a multi-task convolutional neural network, which simultaneously achieves higher accuracy in face recognition tasks and attribute classification tasks. The method includes the following steps: S1, preprocessing the face image sample; S2, extracting attribute features and global face features from the preprocessed face image sample through the designed multi-task convolutional neural network model; S3, Calculate the correlation between different attributes and face recognition tasks based on the attention mechanism, and integrate attribute features into face features according to the correlation; S4, multi-task convolutional neural network model simultaneously performs face recognition tasks and attribute classification tasks, And train the optimization model by calculating the loss; S5, use the optimized model to simultaneously perform attribute classification and face recognition tasks on the input face image.

Description

technical field [0001] The invention relates to computer vision technology, in particular to a face recognition and attribute classification method based on a multi-task convolutional neural network. Background technique [0002] Face recognition is a computer-based biometric identification technology. The common feature of other biometric identification technologies such as fingerprint identification, iris identification, and voiceprint identification is that the features used for identification are unique and not easily changed. To ensure that the information is not easily forged and not easily confused. Face recognition has the characteristics of visual perception close to the face, easy to obtain and can be retrieved after the event, so it has been widely used in various fields such as security monitoring, online payment, access control and attendance. Attributes serve as an intermediate representation of recognition targets, which provides an abstraction between low-di...

AI Technical Summary

Problems solved by technology

These two methods integrate each attribute into the face recognition task with equal weight, and will add a lot of task-independent noise, making it difficult for the model to achieve high accuracy on multiple tasks at the same time.

Images