Face And Expression Recognition Research Based On Locally Linear Embedding And Extreme Learning Machine

In the field of face recognition,benefit from the rapid development of neural network algorithms,face recognition methods based on neural network algorithms have been continuously proposed.Researchers' in-depth research in the field of face recognition had greatly developed the face recognition methods.Among them,the Locally Linear Embedding(LLE)algorithm had good processing ability for non-linear data;The Extreme Learning Machine(ELM)had succinct theory and simple to implement.In the field of face recognition in recent years,LLE and ELM had attracted more and more scholars' attention.However,some problems were encountered in actual application process.The problems included the large number of pictures,irregular data distribution,susceptible to noise interference and had outliers.And the problem of feature data extracted from different types of sample data may affect the classical ability of the ELM algorithm,leading to the ELM algorithm insufficiently learned.This seriously affects the recognition accuracy and speed in the face recognition algorithm.In order to solve these two problems,the thesis proposes two improved algorithms through research and improvement,which were respectively applied to face recognition and expression recognition in the field of face recognition.1.A face recognition algorithm based on locally linear embedding-extreme learning machine(LLE-ELM).By using the locally linear embedding(LLE)algorithm the nearest neighbors of the feature points were found,then the local weight matrix of the feature points and the neighbor points was constructed,finally the feature matrix of all samples was obtained.This matrix kept the characteristic structure of the original data,reduced the data dimension,and reduced the computational complexity.Then through the extreme learning machine(ELM)algorithm classification.The feature matrix is used as the input data of the ELM mode.The single layer neural network is used to train and learn.The output weights were obtained by solving the locally linear system.Finally,the face recognition was realized.The experimental results showed that the face recognition model based on LLE-ELM algorithm could improve the face recognition accuracy and speed.2.A facial expression recognition algorithm based on within-class locally linear embedding extreme learning machine(WCLLELM).Because the traditional extreme learning machine algorithm was trained on a small number of label samples,the problem of insufficient learning process was prone to occur.And it was overlooked intrinsic geometric information of the sample in learning process,and its ability to recognize facial expressions was affected.This thesis introduced classification information and used LLE algorithm to construct the within-class local weight matrix of the sample data.Then it could obtain the within-class divergence feature matrix of the sample.In this way,it was guaranteed that the sample of the same type were more concentrated.This way reduced the amount of data while maintaining the essential structure of the data and the discrimination information of the same type of data.Finally,it was improved the overall classification performance of ELM.The experimental results showed that the facial expressions based on WCLLELM algorithm could improve the recognition accuracy and speed.