Face Recognition System Based On DM642

As one of the hot spots in AI (Artificial Intelligence) and Pattern Recognition area, the face recognition could be used in people identification, security supervising in public area, searching in the image database, improvement of the interaction between computers and human-beings and so on. Compared with other biological identification methods, such as the detection of iris, fingerprints and DNA, the face recognition is more fast and convenient, In recent years, the DSP processor has been widely used in the development of face recognition and other types of image processing systems due to its remarkable capability on data analysis and signal processing, as well as the comparing low cost.In this thesis, a face recognition and network video supervising system has been established based on the ICETEK-DM642-PCI development board produced by RuiTai Company. On the foundation of the standard human face images captured by cameras, the face recognition operation is well accomplished. While there is no recognition request from the control terminal on PC, the real-time video images which are compressed in the H.263 format will be transmitted to PC for the security supervising. The major functions realized in this system are as following:(1) the capture of human face images in the size of 92×112 by a PAL camera and the configuration of "Video Port and the AD decoding unit on DSP development board; (2) the pre-processing of the raw face images and the transmission of the processed images to PC; (3) the transmission of the train image set from PC to DSP; (4):the feature extracting and face matching by the improved modular PCA and fisherface method; (5) the software platform on PC programmed by VB for controlling the transmission of instructions and data between PC and DSP.According to the test results, since the strong data processing capability of DM642, the supervising video images can be transmitted smoothly. Limited to the hardware conditions, the number of training image is limited as well, which has to be improved further. Besides, when the MPCA algorithm is running on the DSP embedding system, based on the current hardware and software conditions, the recognition accuracy rate is up to 90%and stable around 85% in average with the capture condition is adjusted. This result is closed to the theoretical simulation in MATLAB. Thirdly, the system resource occupying rate is 90% at most. The entire system is running stably.