A Face Recognition Method Based on Improved Incremental Non-negative Matrix Factorization

Abstract

The invention discloses a face recognition method based on improved incremental non-negative matrix decomposition, which adopts an incremental non-negative matrix solution algorithm based on Fisher discriminant analysis. The algorithm uses the prior information of the initial sample training, assigns the column vector of the new coefficient to the mean vector of the corresponding category through the index matrix, and improves the convergence speed of the algorithm; in addition, the INMF algorithm is improved to a batch incremental learning algorithm , use the index matrix for initialization assignment, and impose the constraints of maximum inter-class scatter and minimum in-degree on the objective function, so as to obtain the best subspace projection. This solution has a higher recognition rate and a faster recognition speed, and is suitable for occasions such as face recognition.

Description

technical field [0001] The invention relates to the field of computer face recognition, in particular to a face recognition method based on improved incremental non-negative matrix decomposition. Background technique [0002] In the past two decades, face recognition technology, as an efficient biometric identification technology, has been increasingly valued by academia and industry. The key to face recognition research is feature extraction, which can be based on global features or local features. [0003] Non-negative matrix factorization (NMF) is a matrix decomposition method under the constraint that the matrix elements are all non-negative numbers. NMF is a local feature extraction method, which represents the face as a linear combination of base images, and the base image represents the local features of the face such as eyes, nose, mouth, etc., which is in line with the concept of human thinking that the parts form the whole. This method was first proposed by Lee e...

AI Technical Summary

Problems solved by technology

In addition, when the one-time newly added training samples reach a certain scale, the time required for training even exceeds that of the NMF algorithm.

[0007] 2) The INMF algorithm is an unsupervised algorithm, which fails to effectively use the category information of the training samples, resulting in a high degree of inter-class confusion in the subspace obtained after dimensionality reduction, which is not conducive to the discrimination and classification of data

[0008] 3) Each time the INMF algorithm adds new samples, the initialization of its coefficient matrix H is randomly assigned, which is not conducive to the rapid decline of the gradient of the objective function in the subsequent iterative solution process, and the convergence speed is slow

Images