SNN-based edge feature extraction and facial expression recognition method

Abstract

The invention discloses an SNN-based edge feature extraction and facial expression recognition method, and the method comprises the steps: 1), extracting the saliency features of a face image, and 2), extracting the features of the face image in an eye region and a mouth and nose region through a multi-task convolutional neural network; 3) carrying out delay coding on the central concave area extracted in the step 2), and realizing pulse sequence conversion of the image features in the time dimension; 4) extracting edge features of the pulse sequence through SNN; 5) predicting expression categories of the eye area and the mouth area through an echo state network; 6) judging whether the predicted expression categories of the eye region and the mouth region are the same, and if so, outputting an expression recognition result; and if not, calculating a competition value H in combination with a feature face algorithm, and outputting an expression recognition result. According to the method, key texture information of facial expressions can be fully learned, high discrimination is achieved, and high recognition precision can be better guaranteed under the condition that the number of training sets is small.

Description

technical field [0001] The invention relates to the technical field of expression recognition, in particular to a method for edge feature extraction and facial expression recognition based on SNN. Background technique [0002] Facial expressions play an important role in human communication. The first research on expression recognition mainly focused on geometric topology. These methods seem to be popular and intuitive, and can reduce the dimension of input data, but usually require researchers to manually label in advance, which is very time-consuming and labor-intensive. Since the 21st century, people have begun to use deep learning methods to solve problems related to facial expression recognition. However, this type of artificial neural network has the following inherent limitations: [0003] First, the deep neural network widely used in the field of AI is not explanatory in terms of feature understanding, and cannot simulate the pulse discharge pattern of real biologi...

AI Technical Summary

Problems solved by technology

[0005] In view of this, the purpose of the present invention is a kind of SNN-based edge feature extraction and facial expression recognition method, to solve the technical problem of poor biological interpretation of existing artificial neural network algorithms, and to solve the problem of ensuring that the number of training sets is small. Technical issues with higher recognition accuracy

Images