| With the development of science and technology in the 21 st century,face recognition technology has changed our way of life.However,face recognition algorithm is susceptible to complex environment such as illumination intensity,occlusion,and face shooting Angle,which reduces the efficiency of face recognition.The traditional face recognition system is mainly based on PC,which is not only bulky,but also inconvenient to operate.With the development of embedded technology,face recognition technology has been widely used in embedded field.Small size,low power consumption,high integration and portability promote the development of embedded technology.At present,a variety of embedded products have appeared,such as switches,medical instruments,automobile electronics,etc.Therefore,embedded technology has important research value and significance.In order to solve the above problems,this paper studies and analyzes the existing face recognition algorithms,and summarizes the advantages and disadvantages of face recognition algorithms.This paper focuses on the face feature extraction method of 2D-Gabor and LBPH algorithms.Since a lot of redundant information will be generated when 2D-GABor extracts face features,this paper proposes to now weight,reduce dimension and improve processing of 2D-Gabor,and then fuse it with LBPH algorithm for face recognition.The simulation results show that the improved 2D-Gabor and LBPH fusion algorithms have higher recognition rate and robustness than traditional algorithms in complex environments.Finally,the improved LBPH algorithm is transplanted to the embedded development platform to realize embedded face recognition.The system USES Exynos4412 processor to build a client system,which is mainly used for face image acquisition and detection.Linux system is used as the server for face feature recognition,and C++ language is used to write the core algorithm,so as to achieve stable communication between the client and the server.The experimental results show that the embedded face recognition system has the characteristics of high recognition rate,real-time performance and strong anti-interference performance.The main contents of this paper are as follows:(1)Face recognition algorithm is studied and implemented,including histogram equalization algorithm,median filtering algorithm,scale normalization algorithm,Ada Boost face detection algorithm,PCA face recognition algorithm,LDA face recognition algorithm and LBPH face recognition algorithm.(2)An improved LBPH face recognition algorithm,which combined multi-scale 2D Gabor with LBPH algorithm,is proposed.Experiments indicate that the recognition accuracy of the improved algorithm is higher than that of the traditional algorithm.(3)The software and hardware system framework are built.The Samsung Exynos4412 development board based on the Crotex-A9 architecture is selected as the hardware platform for the hardware part.The 5-megapixel USB camera is used to collect face images.For the software,Ubuntu 12.04 is selected as the development platform for PC.Boot program Uboot,Linux operating system kernel,computer vision library Open CV,QT graphical interface library and other port work are finished.(4)Face recognition based on improved LBPH algorithm is realized on embedded development platform.The verification tests on face recognition system are conducted to identify the performance indexes.Experimental results show that the face recognition system has a good performance on the recognition speed and accuracy.It can satisfy people's needs and has research significance and application value. |
