Face Recognition Based On Improved LBP And 2DLDA Algorithm

Compared with other biometric methods,face recognition has the characteristics of no invasive,no contact,no need for passive cooperation and scalability,and is widely used in various fields.Face recognition develops rapidly and the recognition technology is relatively mature.However,face recognition will be affected by factors such as illumination,expression and posture.There are still many difficulties that need to be overcome.The Local Binary Pattern(LBP)algorithm has the characteristics of being insensitive to illumination and easy to understand,and has high research value in face recognition technology.This paper is based on the LBP algorithm.The main work and innovations are as follows:(1)In the calculation process of LBP algorithm,only the relationship between the gray value of the neighborhood and the central pixel is considered,and the correlation between the gray level mean of the image pixel and the row,column and diagonal is neglected,which is susceptible to noise and edge points.This paper proposes a Multiple Uniform Local Binary Patterns(MULBP)algorithm.Firstly,the sum of the gray value of the center pixel and the gray value of the row,column and diagonal pixels is calculated,and then the mean value is obtained separately,and then compared with the image gray mean value to obtain the binary code of the image,and the histogram feature is extracted.Experiments show that the influence of noise and edge points on the recognition results can be eliminated to some extent(2)Two Dimensional Linear Discriminant Analysis(2DLDA)algorithm describes the global features of the image and is sensitive to illumination.If the face image is directly analyzed to extract global features,local information is easily lost Aiming at the shortcomings of 2DLDA algorithm,this paper further proposes an algorithm based on MULBP and 2DLDA.Firstly,the face image is segmented,the MULBP algorithm which is not sensitive to illumination is selected to extract local features,the 2D feature matrix is constructed,and then the 2DLDA algorithm is used to find the optimal.The projection matrix,the classifier selects the commonly used nearest neighbor classification algorithm for experimental comparison and verification.The experimental results show that the MULBP algorithm and the 2DLDA algorithm are complementary,which significantly improves the recognition effect of the 2DLDA algorithm and is robust to noise and illumination effects.(3)Since the nearest neighbor classification algorithm only considers the distance between the test sample and the single sample in the training sample set,it is easy to cause misclassification.Collaborative representation classification algorithm linearly represents test samples with all training samples.From the perspective of global classification,a collaborative representation classification algorithm based on MULBP and 2DLDA is proposed,and compared with MULBP and 2DLDA algorithms.The experimental results show that the classification effect of the algorithm is more it is good.Figure[28]table[5]reference[59]...