Method for face recognition through classification layer supervision based on Sigmoid function

Abstract

The invention discloses a method for face recognition through classification layer supervision based on a Sigmoid function, and belongs to the field of computer vision. The invention provides a loss function which can be used for training a face recognition deep neural network based on two aspects of a Sigmoid nonlinear function and Softmax classification layer supervision. The method is suitable for current face recognition network training based on the deep neural network. Experiments on a public data set show that a face recognition network obtained by using the method as loss function training has high face recognition accuracy, and also has high face recognition capability in extreme scenes such as side face shooting and strong and weak light irradiation.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a method for face recognition based on Sigmoid function-based classification layer supervision. Background technique [0002] Since the 21st century, face recognition technology has been a hot research issue in the field of computer vision, and it is widely used in security, people's livelihood, transportation and other fields. In the current research on face recognition technology based on deep neural network, network structure optimization and loss function optimization are two main research directions. The optimization of the network structure mainly focuses on how the feature extraction network in the face recognition task can extract more effective face features. These features include low-level features such as contours and colors, as well as high-level features rich in semantic information. The optimization of the network structure can improve the accuracy of face recogniti...

AI Technical Summary

Problems solved by technology

[0004] The image classification network trained using the Softmax classification layer supervision can accurately classify thousands of natural images, but it cannot be directly applied to the face recognition network.

The Softmax loss function only pays attention to the correctly classified categories, and does not optimize the wrongly classified categories. Therefore, the intra-class distance of the features trained using the Softmax loss function is not compact enough, and the inter-class distance is not far enough.

Images