Multi-modal face recognition method based on attention mechanism

Abstract

The invention is applied to the field of pattern recognition, and provides a multi-modal face recognition method based on an attention mechanism, a network model afr_net constructs the attention mechanism by adopting a CBAM and an SAVM. A space and channel attention module CBAM is added in each block of ResNet18. ResNet18 combined with an attention mechanism is utilized to establish RGB, depth and branches of fusion modes of the RGB and the depth, so that features of three modes are obtained, then the features of the three modes are fused and input into a sharing layer. The feature vectors are obtained through a vectorization module SAVM based on the space attention mechanism and a full connection layer. The method not only overcomes the defects of a traditional two-dimensional face recognition method, but also effectively fuses RGB and depth modals, and enhances the RGB-D face recognition capability.

Description

technical field [0001] The invention relates to the fields of machine deep learning and image processing and recognition, in particular to a multimodal face recognition method based on an attention mechanism. Background technique [0002] The technical research of face recognition began in the 1960s, and gradually became a hot topic in the field of computer vision. Recently, with the rapid progress of deep learning technology and the open source of a large number of two-dimensional face data sets, two-dimensional face recognition It was a huge success. Since AlexNet was proposed and cited in 2012, most face recognition models have adopted a deep learning strategy with CNN as the backbone. In 2015, Google proposed the FaceNet model, which achieved an accuracy of 99.63% on the LFW benchmark, and its performance surpassed that of humans. Most of these two-dimensional face recognition progresses use two-dimensional images (RGB), and two-dimensional RGB images contain limited f...

AI Technical Summary

Problems solved by technology

[0004] Most of the existing 3D face recognition uses data fusion, feature fusion or fractional fusion to improve the performance of the 3D face recognition model, such as: Chinese patent" 201911397767.1" disclosed "a 3D face recognition method based on the fusion of multiple data types", "201410343015.8" disclosed "3D face recognition method based on feature points" and "202011000990.0" disclosed "a Low-precision 3D face recognition method based on depth map quality enhancement", etc., the recognition accuracy of these existing technologies is low, which cannot meet the existing needs

Images