Facial expression recognition method based on feature fusion and BP neural network

Abstract

The invention discloses a facial expression recognition method based on feature fusion and a BP neural network. The method comprises the steps of firstly, cutting a background area of an original image in an expression database, and acquiring a picture with only a face part through cutting and graying; obtaining a feature value according to the covariance matrix by using the facial expression features after picture preprocessing to realize data dimension reduction; and performing serial feature fusion on the preprocessed features and the features subjected to dimension reduction, training feature vectors subjected to feature fusion through a neural network to obtain a classification model, and predicting and identifying expressions through the established classification model. The information amount of the image is rich, the key face features of the core can be highlighted. Amodel is simple, a large amount of data is not needed, and several types of basic expressions can be effectivelyrecognized by considering the global features and the face features subjected to dimension reduction.

Description

technical field [0001] The present invention relates to a kind of pattern recognition technical field, more specifically, relate to a kind of facial expression recognition method based on feature fusion and BP neural network. Background technique [0002] Facial expressions are one of the important ways we express emotions, and expression recognition is also one of the most powerful and challenging tasks in social communication. Today's face detection technology has matured in recent years, and facial expression recognition is also in the stage of rapid development and research. Expression recognition involves many disciplines, such as artificial intelligence, pattern recognition, physiology and medicine. Expression recognition also has broad development prospects in many fields such as psychological research, vehicle safety driving, clinical medicine, and human-computer interaction equipment. [0003] Expression recognition can generally be divided into three steps: collec...

AI Technical Summary

Problems solved by technology

Scholars mostly study under the single feature of feature extraction, and a single feature often cannot accurately describe the content of the image, and cannot describe the subtle characteristics and changes in the expression.

How to obtain a better description of the key features in the representation of the image, for the original features, its information is relatively rich, but the disadvantage is that the data is redundant. In addition to the main key features, there are many other interference factors such as Beijing and noise, which will lead to focus not outstanding

However, the features after PCA feature extraction and dimensionality reduction can highlight the key face features, but some subtle face information is lost during the transformation process.

[0004] Chinese Patent Publication CN109858467A disclosed on June 07, 2019 "a face recognition method and device based on key point regional feature fusion", which extracts The fusion of features is to fuse several key points of the face. The model used is more complex and requires a large amount of data training, and it is easy to ignore the relationship between the whole and the part.

In the process of facial expression recognition, it is often easily affected by changes in face, age, gender, race, occlusion, etc., while the traditional method recognizes a single feature. This disadvantage is that the information of the whole face cannot be trained as a feature in the network