Binary-channel convolutional neural network for facial expression recognition

Abstract

The invention discloses a method for recognizing human facial expression by virtue of a binary-channel convolutional neural network. According to the method, firstly, pretreatments are carried out aiming at different input signals and include human face detection, rotation correction, downsampling and data sample expansion (if RGB images are input, the RGB images are grayed so as to decrease computation complexity), so that the human face detection precision is improved; secondly, LBP images corresponding to sample-expanded gray images are calculated so as to form a binary-channel sample set for subsequent model training and testing; a binary channel feature extraction network (Binary Channel-Feature Extraction Network, BC-FEN) is utilized for effectively extracting overall and local features of a human face image; and finally, a weighted merge classify network (Weighted Merge Classify Network, WMCN) is utilized for finishing the feature fusion and expression classification of the human face image, so that the human face expression recognition precision is improved.

Description

technical field [0001] The invention belongs to the field of intelligent monitoring, in particular to a dual-channel convolutional neural network for facial expression recognition. Background technique [0002] Accompanied facial expression recognition refers to the use of computer vision technology to predict expressions from human face images. It plays a great role in revealing people's intentions, emotions and other internal states. It is an important tool for machines to perceive human emotional changes and communicate with humans. It has been widely used in human-computer interaction, health monitoring, and assisted driving. [0003] The facial expression recognition process includes image preprocessing, facial feature extraction, and expression classification. Face detection is usually implemented using cascaded classifiers, such as the currently popular Viola-Jones face detection framework. After the face is detected, it can use feature points such as eyes and mouth...

AI Technical Summary

Problems solved by technology

[0004] In recent years, facial expression recognition has made great progress. However, complex factors such as illumination changes, partial occlusion, and facial rotation usually affect the results of face detection, thereby reducing the accuracy of expression recognition.

Even if the face is detected accurately, facial expression recognition is a very challenging task. The difficulty lies in: (1) facial images of the same expression may vary from person to person; (2) the difference between different expressions of the same subject may not be obvious ; (3) The intensity of the same expression may lead to differences in facial images

Images