Study On Standard Model Features For Face Recognition

As a typical biological recognition technology, face recognition technology has a wide range of applications in business, judicial, monitoring and video retrieval fields. Compared with other human biological characteristics, the face data has some direct, friendly, convenient advantages. But the face recognition algorithm performance is not good enough at present, so that the performance of the machine vision is far less than human itself. Based on the biologically-inspired knowledge to model the biological visual nervous system process in visual perception task, we present a model based on the visual perception mechanism for face recognition.The paper firstly introduces the relevant knowledge of face recognition and bio-vision, then presents an improved face recognition algorithm using standard model feature. The face recognition algorithm based on subspace, principal component analysis (PCA), linear discriminant analysis (LDA) and nonnegative matrices decomposition (NMF), are choosed as the benchmark algorithm for experiments. Experimental results show that the new algorithm is better than the above three algorithm in robustness and accuracy. At the same time, the introduction of receptive field is successful and plays a positive impact on the improvement of recognition performance.In the stability evaluation of algorithms, we use the Bootstrap methods to evaluate all these algorithms. We conduct an experiment of 50 times random sampling and another of 200 times. Through comparing the experimental results, we can see the new algorithm has the better stability, and when the database is simpler, algorithm based on LDA is also praiseworthy in accuracy and stability.