Discriminative feature learning method and system for micro-expression recognition

Abstract

The invention discloses a discriminative feature learning method and system for micro-expression recognition. The method comprises the following steps: firstly, extracting a start frame and a peak frame in a micro-expression video sequence, preprocessing the start frame and the peak frame, and further calculating optical flow information between the peak frame and the start frame to obtain an optical flow graph; then selecting an image of which the expression category is different from that of the peak frame from the common expression image library, cutting the image, and replacing a corresponding region of the peak frame image with an image block obtained by cutting to obtain a composite image; constructing a double-flow convolutional neural network model based on a class activation graph attention mechanism, inputting the optical flow graph and the composite image into two branches of a double-flow convolutional neural network respectively, and training the model; and finally, extracting features with strong discriminability from an input video sequence by using the trained model for micro-expression classification and recognition. The method can effectively prevent the model from being over-fitted, enables the model to learn the micro-expression features with high discriminability, and improves the accuracy of micro-expression recognition.

Description

technical field [0001] The invention relates to a discriminative feature learning method and system for micro-expression recognition, belonging to the field of micro-expression recognition and artificial intelligence. Background technique [0002] Expression is a non-verbal behavior for humans to express their emotions, and it is also an important way for robots to intelligently understand human emotions. General facial expressions are shown by people when their emotional expression is not suppressed, and the range of facial movements is often large and lasts for a long time. But in some cases, people will deliberately suppress and hide their emotions, and these suppressed emotions will be expressed spontaneously through extremely fast facial expressions. This type of expression is called micro-expression. The duration of micro-expressions is extremely short, less than 0.2 seconds, and the changes in facial movements are so subtle that the accuracy of human recognition of m...

AI Technical Summary

Problems solved by technology

[0007] Purpose of the invention: In view of the problems of model overfitting and insufficient extraction of micro-expression discriminative features in the micro-expression recognition method based on convolutional neural network feature extraction, the present invention proposes a discriminative feature learning method for micro-expression recognition Method and system, using two-stream convolutional neural network to extract discriminative features of micro-expression video sequences, in order to make the model not only pay attention to facial changes but also pay attention to other facial changes compared to obvious areas (such as mouth, eyes, etc.) For intersecting tiny areas, the present invention uses the synthetic image obtained from the ordinary expression image and the micro-expression peak frame image and the optical flow graph calculated from the micro-expression peak frame and the initial frame to train the dual-stream convolutional neural network

Images