Expression recognition method based on loss function integration and coarse and fine hierarchical convolutional neural network

Abstract

The invention discloses an expression recognition method based on loss function integration and a coarse and fine hierarchical convolutional neural network, and the method comprises the steps: carrying out the improvement from the angle of a loss function for solving a problem that the classification recognition accuracy is affected because the inter-class distance of expression features extracted by the convolutional neural network is too small and the intra-class distance is too large, introducing other four loss functions to replace a common Softmax loss function so as to enlarge the inter-class distance of expression features and reduce the intra-class distance; the invention provides an expression recognition method based on the coarse and fine hierarchical convolutional neural network, and aims at the problem of classification and confusion of several types of expressions in expression recognition. In order to unify expression recognition task functions, an expression recognition system based on the convolutional neural network is designed and developed. According to the method, the expression recognition accuracy of different loss functions can be improved, the recognition accuracy of an integrated network is improved, and the recognition accuracy of easy-to-confuse type expressions is improved.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to an expression recognition method based on loss function integration and coarse and subdivision convolutional neural network Background technique [0002] There are various ways for humans to express emotions, such as changes in posture, severity of speech, facial expressions, etc. Among them, facial expressions play an important role in emotional expression. With the advancement of technology and the continuous improvement of living standards, people have higher expectations and requirements for intelligent life, and the research on facial expression recognition continues to heat up. For computers, it is becoming more and more important to give correct emotion classification results based on human expressions. As early as the twentieth century, Ekman and Friesen began the research on human expressions and defined six basic expressions of cross-ethnic groups, including a...

AI Technical Summary

Problems solved by technology

[0004] Traditionally, in order to overcome the problem that the distance between classes is too small and the distance within a class is too large to affect the accuracy of classification and recognition in the expression features extracted by convolutional neural networks, the present invention provides a convolutional neural network based on loss function integration and coarse and subdivision classification. The network expression recognition method can improve the expression recognition accuracy of different loss functions, improve the recognition accuracy of the integrated network, and improve the recognition accuracy of confusing categories of expressions

Images