Research On Feature Extraction Of Face Image Based On Gabor-LDA And K-Algorithm Of Correlation Coefficient

In the field of biometrics,face recognition is considered to be the most commonly used modality.As the face recognition technology has a good high popularity,non-contact,non-disturbing and other advantages,so in many areas such as criminal investigation,suspect tracking,access control systems,monitoring and information security has been successfully applied,especially in recent years,Face recognition technology has been a wide range of attention and research.Face recognition technology mainly includes three aspects:face detection,feature extraction and classifier design,this paper mainly from the face image feature extraction and classifier design two aspects of research.Because the actual environment(such as monitoring)often encounter a large number of low-resolution face images,if the use of traditional methods to extract image features directly,the results often reach the satisfaction of people,so in order to overcome the low resolution of this factor on the people Face recognition results,this paper presents a Gabor wavelet feature fusion method based on a variety of low resolution.We decompose low-resolution images into three different scales of different resolution images to construct multi-scale multi-resolution fusion image sequences.Each image sequence uses Gabor wavelet to extract face features,and then uses linear discriminant analysis to obtain high-dimensional Characteristics of the initial dimensionality reduction feature extraction.Then,the multi-resolution images of three different scales after the feature extraction are merged,and the linear dimension analysis is used to reduce the redundancy,and then the K-nearest neighbor classifier is used for classification prediction.Because the difference between face features is very small,if we use K near neighbor classifier for face image classification,the effect is often very poor,so this paper also proposed based on the correlation coefficient K nearest neighbor classifier classification,the use of correlation coefficient As a criterion for the distance between the subjects.Finally,the paper uses ORL,YALE and AR to compare the three standard face databases.The experimental results show that the multi-structure feature fusion method based on low resolution can extract more and more effective face features.At the same time,The application of classifier in face recognition can improve the accuracy of face recognition effectively,and it has certain theoretical value and practical value.