Micro-Expression Recognition Based On Mean Gray Local Ternary Patterns And ELM

In recent years, research on micro-expression has made a rapid development, and micro-expression recognition is an emerging topic in the field of human-computer interaction. It is a very fast expression which people try to hide the true expression of the feelings, and duration time is only 1/25 second to 1/5 second, expressing the six basic facial expressions. By analyzing micro-expression, the true feelings which people try to hide can be found. It is widely used in clinical, justice, security and other fields. So it has a very important practical significance to study micro-expression.Many steps have been included in the micro-expression classification and recognition. Feature extraction and expression classification are the key technologies of the micro-expression recognition. In face expression recognition field, local binary pattern(LBP) and local ternary pattern(LTP) are described as the local texture feature extraction operators that have been widely used. These operators can be also used in the micro-expression recognition field. Extreme learning machine(ELM) based on a single hidden layer feed forward neural network is a new algorithm to classify expressions, and it has the advantages of fast and easy operation. Considering the shortcomings of traditional feature extraction methods, firstly, a new algorithm called mean gray local ternary pattern(MG-LTP) based on the average gray of LTP for expression feature extraction has been proposed in this paper; secondly, principal component analysis(PCA) has been used for data dimensionality reduction; then, extreme learning machine(ELM) has been used as a classifier for feature classification; finally, expression recognition experiments on JAFFE database and micro-expression recognition experiments on CASME database have been accomplished. Compared with the traditional image recognition methods, the methods used in this paper get better results.