A facial expression recognition method based on multi-scale feature extraction and global feature fusion is proposed

Abstract

The invention discloses a face expression recognition method of multi-scale feature extraction and global feature fusion. A human face expression data set is selected as raw data, and the raw data isdivided into training set data and test set data. A TensorFlow artificial intelligence learning system is used to construct a convolution neural network with multi-scale feature extraction and globalfeature fusion. The convolution neural network reads the training set data, pretreats the training set data and performs model training, then reads the test set data, recognizes the expression types of each expression in the test set data in turn, and calculates the average accuracy rate and average F1 of all expressions after completing the recognition of all expressions. Score indicators, the final completion of the process of facial expression recognition. The invention has the advantages of high recognition speed under the condition of ensuring high recognition accuracy, and can be adaptedto multiple illumination environments with strong robustness, so as to effectively meet the practical application requirements.

Description

technical field [0001] The invention relates to a facial expression recognition method, in particular to a facial expression recognition method of multi-scale feature extraction and global feature fusion. Background technique [0002] In people's ordinary communication, facial expression is one of the most important ways. As a carrier of information, it contains a lot of information that cannot be expressed in language, which we call non-verbal information. Since faces can be directly observed in communication, facial expressions can convey emotional information in a more intuitive way. In recent years, emotion analysis has gradually attracted the attention of many researchers, and facial expression recognition is an important part of emotion analysis, and it has also been developed rapidly, especially in intelligent human-computer interaction. To infer people's emotional information and perform emotional imitation. [0003] The steps of the current facial expression recog...

AI Technical Summary

Problems solved by technology

However, traditional computational vision methods also have deficiencies: 1. Man-made design rules for feature extraction, so that the final extracted features contain some subjective factors, which cannot truly and objectively reflect the true distribution of data; 2. Feature extraction methods with excellent performance often have a large amount of calculation , it is difficult to realize real-time recognition on limited hardware resources

However, in these existing recognition methods, the recognition speed is relatively slow when the facial expression recognition accuracy is high; and the facial expression recognition accuracy is low when the recognition speed is fast.

Images