Research On Micro Expression Recognition Based On Transfer Learning

Micro-expressions are very brief facial expressions that people unconsciously produce in a stressful environment or when they try to hide their true emotions,which can reveal the true mental activities of people.Micro-expressions have potentially great application value in many fields such as national security,medical care,educational psychology,and political psychology.The frequency of micro expressions is low and the duration is very short,which is generally not observed by human eyes.People turn to computers to study micro-expression.With the development of machine learning,especially the continuous production of new algorithms in the fields of computer vision and pattern recognition,the development of micro-expression research methods has also continued to expand.Currently there are few micro-expression samples because of less micro-expression databases.Building a micro-expression recognition model directly using these samples can't reach good experimental results.When observing the database,it is found that the amount of micro-expression data is small,but relatively speaking there is more macro-expression data.Different types of expressions include both macro-expressions and micro-expressions.There are similar characteristics between macro-expressions and micro-expressions.For example,there are anger characteristics between the macro-expressions of anger and the micro-expressions of anger.Transfer learning uses the similarity of macro-expressions and micro-expressions,and at the feature level,it obtains features similar to micro-expressions in macro-expressions,which is equivalent to expanding the characteristics of micro-expressions.In response to the above shortcoming and idea,this thesis designs a micro-expression recognition method combined with transfer learning.The main work is as follows:(1)Realized new idea.We innovatively proposed an idea using the transfer learning method to obtain similar features in the macro-expressions to assist in the training of the micro-expression recognition model.We used the existing DDC(Deep Domain Confusion)method and Deep CORAL(Correlation Alignment for Deep Domain Adaptation)in transfer learning method to build a micro-expression recognition model.(2)Improved the Deep CORAL model which achieves better results.Based on the Deep CORAL method of transfer learning to establish the micro-expression recognition model,we combined with the deep transfer learning theory to improve the Deep CORAL model,the recognition accuracy of the model is improved.Compared with the latest algorithms at home and abroad,it is proved that this improved model has achieved better micro-expression recognition results...