Facial emotion recognition method based on depth sparse self-encoding network
A sparse self-encoding, facial emotion technology, applied in neural learning methods, character and pattern recognition, biological neural network models, etc., can solve the problems of local optimal gradient, dispersion, etc. Effects of Local Extremum and Gradient Diffusion Problems
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Publication Date
- 2017-03-15
- Estimated Expiration
- Not applicable · inactive patent
Smart Images

Figure 1 
Figure 2 
Figure 3
Abstract
Description
technical field
[0001] The invention relates to a face emotion recognition method based on a deep sparse self-encoding network, and belongs to the technical field of pattern recognition. Background technique
[0002] With the rapid development of theories and technologies such as human-computer interaction and affective computing, people generally hope that robots have the ability to recognize, understand and generate human emotions, so as to achieve harmonious, friendly and smooth human-computer communication. Due to the complexity between the diversity of human emotions and the corresponding behaviors, current human-computer interaction still faces some difficult problems in the field of affective computing (including the ability to recognize, understand, and express emotions). Research on emotion recognition based on facial expressions, speech, gestures, physiological signals and other information has become the focus of human-computer interaction. Facial expression reco...
Examples
Embodiment Construction
[0042] The present invention will be further described below in conjunction with drawings and embodiments.
[0043] The present invention provides a face emotion recognition method based on deep sparse self-encoding network, referring to figure 1, by constructing a deep sparse autoencoder network to learn facial expression features and using a Softmax classifier to perform emotion recognition on expressions. First, use the restricted Boltzmann machine to perform layer-by-layer greedy pre-training to obtain the initial weight matrix of the network, expand the model to generate the "encoding" network and "decoding" network, and then build a Softmax classifier on the top of the model and train it. The gradient descent method is used to find the optimal model parameters, and finally the entire network including the Softmax classifier is regarded as a model, and the backpropagation algorithm and the gradient descent method are used to fine-tune the network weights to achieve the gl...