Abstract: In order to improve the vehicle face recognition accuracy in traffic videos under various illumination conditions， a vehicle face recognition algorithm based on improved nonnegative matrix factorization(NMF)was proposed. The vehicle face image and license plate information were acquired after image preprocessing. The original feature of vehicle face image was extracted adaptively based on the special illumination condition. For the importance variation of different pixels in vehicle face image， a feature dimension reduction based on weighted and sparse NMF(WSNMF)was established. The vehicle legality can be defined by determining the similarity of features and the consistency of license plates. The experimental results showed that the proposed algorithm has better recognition performance， and genuine acceptance rate(GAR)and false rejection rate(FRR)can reach 0.9875 and 0.04， respectively， and meet the real-time requirements.

Key words: vehicle face recognition, video processing, license plate recognition, nonnegative matrix factorization, sparse representation