Research And Implementation On Embedded Face Recognition System Based On Improved LBP Algorithm

Face recognition technology is an important research direction of the development of the emerging science and technology, in recent years, along with the advance of science and technology, many difficult issues in face recognition is defeated, that researcher’s effort is inseparable. In today’s society, the face recognition in various fields of science and business has a huge demand of the market, such as public security, real-time monitoring and e-commerce and other fields. Due to the complexity of human face recognition algorithm, the performance of embedded hardware system to a huge now, for the development of the embedded system because of its portable, mobile, and other fields is able to succeed.For human face feature, it is the contact resistance, good stability, no need to cooperate with wait for a characteristic, that face recognition have topic widely attention. But because the face image will be affected by light, angle, position, adorn article, the influence of such factors as lead to face image recognition precision is on the low side, at the same time, it can make person face recognition algorithm is more complicated.In this paper, the images are normalized with same unified standard, and then decomposed by wavelet transform. The images can effectively be decomposed into low frequency and high frequency information. Wavelet transform can obtain lower dimension image data, filter noise, extract some key information, and improve the performance of the follow-up feature extraction. It also can improve the efficiency and accuracy. Through the study of characteristics of LBP(local binary pattern), we extend LBP features to multi-scale LBP. And the secondary of wavelet decomposition approximation histogram of the image can improve the recognition accuracy. At the same time, the Histogram of Oriented Gradient is used to extract the image characteristics and carry on statistics to calculate gradient amplitude and direction histogram, fusion with multi-scale characteristics of LBP histogram. After training and learning of these face features, the final recognition result can be got.Due to the development of the embedded system technology, its performance was improved. We can use cross-compilation injection related program for embedded systems. This paper adopts Hi3531 embedded processor embedded processing system platform, using the Windows CE as the hosting operating system, with vc++ language to develop the core algorithm, to meet the demand of the run on an embedded system. The experimental results on Ara v. Nefian face library show that the development platform has help to recognition accuracy and efficiency.