Action recognition, liveness detection and model training methods and devices, electronic equipment

By combining channel self-attention mechanism and attention mechanism in category recognition operations, the key information of the target image is focused, which solves the problem of low accuracy of traditional action recognition methods and achieves more efficient facial action category recognition and liveness detection.

CN116152908BActive Publication Date: 2026-06-30MASHANG CONSUMER FINANCE CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
MASHANG CONSUMER FINANCE CO LTD
Filing Date
2022-08-30
Publication Date
2026-06-30

AI Technical Summary

Technical Problem

Traditional motion recognition methods have low accuracy in identifying facial motion categories and cannot meet the requirements for fast and accurate liveness detection.

Method used

A category recognition operation combining channel self-attention mechanism is adopted. By using the attention mechanism to focus on important information and ignore secondary information during feature extraction, the obvious features of the target image are extracted by combining details of different spatial domains on multiple channels, thereby improving the accuracy of predicting action categories.

Benefits of technology

It improves the accuracy of facial action category recognition, simplifies the recognition process, reduces complexity, and enhances the efficiency and security of liveness detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116152908B_ABST
    Figure CN116152908B_ABST
Patent Text Reader

Abstract

This application discloses an action recognition method, a liveness detection method, a model training method, an action recognition device, a liveness detection device, a model training device, an electronic device, and a computer-readable storage medium, relating to the field of image processing technology. The action recognition method is used to identify the action category of a face in a target image. The method includes: performing a category recognition operation on the target image using a channel self-attention mechanism to obtain N predicted action categories corresponding to the target image, where N is a positive integer; and determining the action category of the face in the target image based on the N predicted action categories. By leveraging the attention mechanism's ability to simulate human vision—focusing on important information and ignoring secondary information—the method, through the channel self-attention mechanism, focuses on key information in each channel, highlighting and extracting obvious features. Furthermore, by combining different details in different spatial domains, the receptive field is expanded, thereby improving the accuracy of predicted action categories.
Need to check novelty before this filing date? Find Prior Art