电子器件2025，Vol.48Issue(2)：380-387,8.DOI:10.3969/j.issn.1005-9490.2025.02.022
快速线性注意力机制算法在活体检测中的应用
Application of Fast Linear Attention Algorithm in Face Anti-Spoofing Detection
摘要
Abstract
In order to solve the problem that face recognition system is vulnerable to non-human deception,considering the huge amount of calculation of traditional attention mechanism algorithm,a convolutional neural network silent face liveness detection algorithm based on fast linear attention mechanism is proposed.Firstly,the image data containing the face is extracted from the video,and the data pre-processing is carried out,including the data normalization operation for resisting the influence of illumination and the image rotation op-eration for expanding the dataset.Based on the classical point multiplication attention mechanism,the algorithm linearly optimizes the soft maximization operation,that is,the soft maximization operation is removed,and the original two factors are normalized in their re-spective dimensions.In addition,based on the prior knowledge and multiplication combination law,the matrix factor multiplication order is changed,so that the original complexity is reduced from O(N2)to O(N).Experiments on CASIA-SURF and self-made datasets show that the algorithm can effectively reduce the computational complexity while ensuring the recognition performance.Under the same num-ber of training steps,the training time can be shortened by about 1/8,and as the size of the input image increases,the training time will be further shortened.关键词
计算机视觉/活体检测/卷积神经网络/注意力机制Key words
computer vision/liveness detection/convolutional neural network/attention mechanism分类
信息技术与安全科学引用本文复制引用
李金科,张超,张竞宇,朱益良,王健伟,谢跃..快速线性注意力机制算法在活体检测中的应用[J].电子器件,2025,48(2):380-387,8.基金项目
国家自然科学基金项目(62001215) （62001215）
江苏省大学生创新训练项目(202111276069Y,202211276175H) （202111276069Y,202211276175H）
