Video Stream Face Image Recognition Algorithm Research And DSP Implementation

With the development of Information Science,more and more people tend to use a high-speed computing capability device to process information.At the same time,the recognition of video image will become an intelligent processing.Today,face recognition is a hot research direction.Based on the broad access to the relevant information,this thesis has made a deep research on face image recognition in video.In this thesis,the identification system has been simulated and tested on Matlab.The training samples and the test samples were derived from the ORL face database and the self built face database.The images of self built face database were extracted from the collected videos.The thesis transformed videos into frame images and extracted key frame from videos.After the color detection algorithm was used to locate the human face,this thesis extracted the images containing human faces.After grayed and normalized the extracted images,the thesis generated the face database by using the processed images.The simulation and testing process are as follows:first of all,the images of the face database were reduced by using PCA algorithm.In the second place,this thesis used the improved genetic algorithm to extract the feature of dimensionality reduction images.Finally,the thesis used nearest neighbor classifier and support vector machine classifier to classify the test samples.The test shows that the recognition rates of nearest neighbor classifier are 86.67%and 91.25%based on the ORL face database and the self built face database.The recognition rates of support vector machine are 86.67%and 91.25%.The microprocessor in the thesis is the DaVinci series of DSP.The development platform is ICETEK-DM6437-B.The hardware mainly used CCS3.3 for debugging.The CCS used emulator to establish communication between the computer and the hardware.In this thesis,the hardware implementation of video stream face image recognition system has tested through the software simulation and the recognition function of the software was transplanted to the development board.The main process is as follows:firstly,the Matlab simulation language was changed to the C language for running in the CCS3.3.Then the thesis set the DSP's relevant environmental parameters and profiles by using BIOS.Secondly,the C language program was run and the program was debugged according to recognition requirements.Finally,the generated executable file was downloaded to the development board through the JTAG interface and the recognition result was outputted on the screen.The video stream face image recognition system based on DSP hardware platform of this thesis can reach the average recognition rate over 90%.Through the verification of the identification system,the results indicate that the system can effectively identify face and meet the requirements.