Non-negative matrix factorization face recognition method and system on the basis of kernel machine learning

A non-negative matrix factorization face recognition method and system on the basis of kernel machine learning, the face recognition method comprising: A, representing each preset training sample image as a column vector (S1); B, constructing a symmetric positive semi-definite kernel matrix K xx on the basis of the known kernel function and training sample vectors (S2); C, respectively establishing three objective functions and minimizing the objective functions by means of crossed iteration, and obtaining new features of the training samples in a kernel space and two kernel matrixes associated with the nonlinearly mapped samples (S3); D, through the two kernel matrixes obtained in the leaning stage, projecting test samples to the kernel space to obtain new features of the test samples in the kernel space (S4); and E, using a nearest neighbor method to compare the new features of the test samples with the new features of each type of the preset training sample so as to further classify and recognize the test samples (S5). The method skips over the process of original image learning through directly learning two kernel matrixes K wx and K ww ; the method thus avoids the process of kernel function derivation by changing the learning object, and thus the effect of no limitation for selection of kernel functions is achieved and a universal algorithm for any kernel function is obtained.