High-precision face recognition method based on deep transfer learning

Abstract

The invention discloses a high-precision face recognition method based on deep transfer learning, and the method comprises the steps: firstly carrying out classification training of a deep convolution neural network through a medium-scale face image data set and a SoftMax classification loss function, optimizing model parameters through a gradient descent algorithm, and obtaining an initial facerecognition model; and then replacing a large-scale target data set, loading the first N-1 layers, except the SoftMax classification layer, of the initial face model, training the loaded deep convolutional neural network by using an Arcface loss function, and optimizing model parameters by using gradient descent to obtain a target face recognition model. According to the invention, deep migrationlearning technology is utilized; the problem that the SoftMax loss function is not high in precision is solved, the problem that convergence is difficult due to the fact that the Arcface loss functionis directly used is also solved, the training speed and the recognition precision of the face recognition model are improved, and finally the recognition precision on a public test set LFW is 99.40%.

Description

technical field [0001] The invention relates to the fields of artificial intelligence and image processing, in particular to a high-precision face recognition method based on deep transfer learning. Background technique [0002] Face recognition is macroscopically divided into two categories: face verification and face recognition. Face verification is a 1:1 comparison, that is, to judge whether the people in the two pictures are the same person. Face recognition is a 1:N comparison, which is to judge which person the system is currently seeing is one of the many people it has seen before. The face recognition algorithm has gone through three stages: early algorithms (based on geometric features, based on template matching, etc.), artificial features + classifiers, and deep learning. At present, the deep learning algorithm has become the mainstream algorithm in the field of face recognition, which greatly improves the accuracy of face recognition and promotes the practical...

AI Technical Summary

Problems solved by technology

[0003] At present, there are mainly the following problems in the field of face recognition: it is difficult to balance the recognition accuracy and model training speed

But because these loss functions are more harsh than the SoftMax loss function, the convergence speed is very slow or even difficult to converge

Images