Facial Expression Recognition Technology Research Based On The Texture Characteristics

Facial expression recognition is an important topic of the research field of human-computer interaction,It receives the extensive attention of scholars.Facial expression recognition can be applied to medical treatment,judiciary,remote network teaching,security monitoring,video retrieval and other fields.With the deep exploration of expression research,people have found a form of expression which is difficult to detect: micro-expression.It occurs in very short time and may be ignored easily.It has the strong practical value in clinical medicine,interrogation and business negotiation.It has become an emerging research topic in the field of human-computer interaction in recent years.The local texture feature extraction has been discussed firstly in this paper.Some methods have been proposed on how to extract static texture information in expression and dynamic ones in micro-expression effectively.The main research work is as follows:In order to overcome the low recognition rate of facial expression which is caused by complex background,a new algorithm called Center Symmetrical Ternary Patterns(CSTP)is proposed in this paper.Firstly,the facial expression image is divided into some sub-blocks,CSTP features are extracted from each sub-block,and the histogram statistics of CSTP features in each sub-block is carried out.Then the information entropy of each sub-block can be calculated in order to construct its adaptive weighted coefficients,followed by multiplying the weighted coefficient by the corresponding CSTP histogram of sub-block.The feature vectors of each adaptive weighted sub-block have been scaled up as the final texture features.In addition,HRI-CSLTP algorithm which has outstanding performance in the feature extraction is applied to facial expression recognition field.The operators of level-vertical CS-LTP and diagonal CS-LTP are combined with each other.Both local texture intensity distribution information and gradient information are provided.Finally,Support Vector Machine(SVM)classifier is used for expression classification.The experiments have been done on JAFFE and CMU-AMP database.The better results of CSTP and HRI-CSLTP have been achieved in expression classification compared with four methods: 2DPCA、Gabor+PCA、LBP and CBP.The algorithm of dynamic texture LBP-TOP is used to micro-expression recognition feature extraction before,but micro-expression recognition rate is not high.HRI-CSLTP and CSTP algorithm which we propose are applied to the micro-expression recognition,and two algorithms of CSTP-TOP and HRI-CSLTP-TOP are further proposed in this paper.The local texture features and global texture features fuse perfectly by the information entropy weight of CSTP-TOP.It is good for the adaptive enhancement of image texture features and promoting robust performance of micro-expression recognition.HRI-CSLTP-TOP algorithm combines the local texture intensity distribution and gradient information.The changes of time and space information in micro-expression are also considered at the same time.The experiments have been done on CASME database.The results show that micro-expression average recognition rates of CSTP-TOP and HRI-CSLTP-TOP algorithms are better than those of VLBP and LBP-TOP.