Research On Face Recognition Based On Transfer Learning

In modern social life,the scale of the data and computing resources is growing very quickly.As a kind of automatically,has the learning methods of artificial intelligence,machine learning,which method not only progresses rapidly in theory,but also develops greatly in practice.Traditional machine learning is dependent,and it requires that the generation mechanism of data do not change with the environment.In most machine learning algorithm studies,there are both training data and real test data characteristics that need to be distributed identically.Based on this assumption,machine learning enables the computer to simulate human learning behavior,thus automatically acquiring knowledge and skills through learning,thus improving the performance of oneself.However,in practical applications,such as computer vision,big data analysis,bioinformatics,etc,this assumption is too strict to be realized.How to analyze and excavate large-scale data in non-stationary environment has become one of the most challenging directions in modern machine learning.Transfer learning eased the require that traditional machine learning must have the independent identically distributed character,and adopt corresponding migration study method for goals and tasks of training data and the need for the distribution of different features,in this way,test data can make those samples not enough task classification recognition results are a lot of ascension.It can excavate the same essential characteristics and structure of the interrelated and different fields between the two domains,which makes a supervision information(such as annotation data,etc.)can be directly transfer and reuse in the field.As for studies of human face recognition,we start with three major directions of feature extraction,dimensionality reduction and classifier classification.The extracted face image features are processed,and the multi source feature migration algorithm based on LPP is summarized and summarized,and the feasibility and superiority of this algorithm are verified by experiments.Then,the static face recognition research is extended to the dynamic facial expression recognition.By comparing the experimental results of the three covariance migration algorithms,a kind of algorithm which is most suitable for facial expression recognition is summarized.The summary is based on the advantages of transfer learning,and a large number of experiments are conducted to verify the mobility of different transfer learning algorithms in facial expression recognition.This paper studies the feature extraction algorithm of human face and face expression recognition in different light color conditions.At present,transfer learning is still in a challenging stage.By studying the current situation of transfer learning and comparing different methods based on the transfer learning,we can apply it to face recognition skillfully to achieve a more ideal effect.