Research On The Dynamic ELM Computing Model For Facial Expression Recognition

By analyzing the facial image to recognize the expression state,the Perception of Users' Emotions is helpful to improve the intelligence of human-computer interaction and serve the human better.Considering the dynamics of expression,that is,the generation and disappearance of facial expression are usually a process,the description of facial expression by image sequence can describe the process of facial expression movement effectively,and has more abundant temporal information than static expression image.It is of great theoretical and practical significance to study how to realize the rapid recognition of expression sequences and improve the recognition rate of expression sequences.In order to solve the above problems,this paper integrates the hidden Markov model and the limit learning machine into a two-layer classification structure called dynamic ELM model.The lower-level HMM model is responsible for extracting the temporal information in the expression sequence,transforming the expression feature sequence into a hidden sequence which is more conducive to reveal the state transition rule of the emotion unit by Viterbi algorithm.The upper-level ELM model is responsible for classifying the hidden sequence.The model inherits the ability of HMM to process and transform dynamic information.On the other hand,the model has the advantage of high efficiency of ELM training,which realizes efficient classification of expression sequences.In addition,since the discrete HMM model can not directly deal with the sequence composed of eigenvectors,we need to transform the sequence composed of vectors into the scalar sequence,and improve the K-modes clustering algorithm to solve the problem,because the extracted image features are tagged data This question.Compared with K-modes algorithm,the clustering performance is improved by matching the frequency of the corresponding labels in each cluster to match the clustering centers.The main innovations and work of the paper are as follows:(1)the main facial expression segmentation and Gabor feature extraction.Considering the different roles of different facial regions in facial expression recognition,this paper divides the main regions of facial expression expression,and uses Gabor feature to extract the feature of each facial expression and PCA dimensionality reduction.At the same time,K-means clustering is carried out in each block,and a clustering label is obtained for each block,and the labels of the main regions are merged to form a complete face feature.(2)An improved K-mode clustering model is proposed.In order to solve the problem that the discrete HMM can not deal with the vector sequence,the K-modes algorithm is improved and the comparison experiment is done.The experimental results show that the improved K-modes algorithm has better clustering performance,and the K-Modes algorithm and improved K-modes algorithm are different in the distance between classes and within distances.(3)A dynamic ELM calculation model is proposed.In order to improve the recognition rate of dynamic expression sequences,an improved dynamic Hidden Markov Model(HMM)with dynamic sequence identification method is proposed,and an ELM model is proposed.The underlying model uses Hidden Markov Model(HMM)to realize the transition from the observation sequence which describes the expression feature to the hidden state sequence revealing the change of expression.Then,the expression of the hidden state sequence is classified and identified by the upper ELM model.The experimental results show that the dynamic ELM method is superior to the traditional HMM in recognition rate and recognition speed.