Face feature extraction method, system and device based on feature re-calibration and medium

Abstract

The invention discloses a face feature extraction method, system and device based on feature re-calibration and a medium, and the method comprises the steps: constructing a face feature extraction network which comprises a plurality of feature extraction blocks; obtaining a face image of a to-be-recognized user; and inputting the face image of the user to be recognized into a face feature extraction network, performing preliminary feature extraction through a feature extraction block, and performing compression, screening and re-calibration on the extracted preliminary feature information to obtain face feature information. According to the method, the features can be screened by utilizing the relationship between the feature channels, the effect of re-calibrating the features is achieved,and a better feature extraction effect can be considered while a higher feature extraction speed is maintained.

Description

technical field [0001] The invention relates to a face feature extraction method, system, equipment and medium based on feature recalibration, and belongs to the field of deep learning and image processing. Background technique [0002] Face recognition is a biometric technology that automatically recognizes people based on their facial features (such as statistical or geometric features), also known as face recognition, portrait recognition, facial recognition, face recognition, facial recognition, etc. . Usually what we call face recognition is the abbreviation of identity recognition and verification based on optical face images. [0003] The process of face recognition is mainly divided into three parts: face detection, face feature extraction and face feature comparison. Among them, face feature extraction is the most critical step in face recognition, and its ability directly determines the recognition accuracy and anti-interference ability of the entire algorithm. ...

AI Technical Summary

Problems solved by technology

However, due to the large number of parameters and the deep network, the image data requires a lot of calculations in the process of forward propagation, and the time consumption is serious.

In the process of distributed training, there is a greater demand for communication with the server, more resources are occupied, and the size of the model generated by training is larger, which is not easy to deploy in some mobile devices or devices with small memory

[0007] 2) The lightweight network has good real-time performance, but the feature extraction effect is not ideal

The lightweight face feature extraction network uses many load reduction measures (such as the use of depth separable convolution), so that it takes up less server resources during training, and when it is actually running, it is faster and the model size is smaller, but the lightweight network The feature extraction ability is limited, resulting in worse robustness, and its feature extraction is easily affected by objective factors such as image brightness and blur

[0008] 3) Existing deep networks rarely consider using the relationship between feature channels to improve network feature extraction capabilities

As the core of the convolutional neural network, the convolution kernel is usually regarded as an information aggregate that aggregates spatial information and feature dimension information on the local receptive field. However, only spatial information is used to train a powerful Networking is quite difficult

Images