Research On Micro-expression Recognition Algorithm Based On ECOC And Deep Learning

As we all know,expressions play a very important role in daily communication.So far,facial expression research has become a hot topic in the field of machine vision and pattern recognition.Micro-expressions are an emotional illusion in which people try to hide their true emotions.Because of the true characteristics of micro-expressions,micro-expression recognition is popular in the many fields.However,the microexpression data set has the characteristics of large number of categories,short duration,small amount of data,and inconspicuous facial expression characteristics,which leads to the research of micro-expression recognition is still in its infancy.Traditional error correction output coding algorithms(ECOC)ignore data characteristics when processing micro-expression data sets.This paper proposes an improved error correction output coding matrix based on multiple data complexity measures named VDCECOC,which uses multiple data complexity measures to generate large-scale coding matrices,and according to the characteristics of microexpression In this paper,a feature selection method based on data complexity is proposed to improve the difference of the coding matrix feature space.Experiments show that the algorithm achieves a good classification effect on the micro-expression data set.Deep learning method is also a research method to solve the multi-classification problem,but the recognition effect is not ideal due to the phenomenon of overfitting.This paper proposes an integrated algorithm named as DeepForest&CNN based on DeepForest and Convolutional Neural Network(CNN),which integrates the rerepresentation features generated by deep forest and high-order features generated by convolutional nerves through integrated learning to differentiate the feature space to improve the adaptability of the model to the micro-expression data set.Experiments show that this method further improves the accuracy of micro-expression recognition.