Research And Implement Of Multi-View Face Detection Algorithm Based On DSP

The purpose of the human face detection is to get the location and size of the specific human face by processing the image.Face detection is considered to be the first step of automatic face analysis system.Its speed and accuracy directly affect the overall performance of the system.Thus it is important to investigate the face image processing,machine vision and pattern recognition.With the development of the society,people have higher and higher requirement for the face detection technology.The face detection can be implemented under the condition with the different posture diversity,illumination,complicated background,which becomes a new direction of research in this filed.In this thesis,the context of this study and the importance of human face detection technology is reviewed firstly,and the algorithms widely used in the face detection study are summarized.YCbCr color space is conducted to examine the human facial skin color clustering.Next the detection of the skin color segmentation method is used to scan the human face candidate area and preliminarily identify the face candidates.Given that the test of skin color shows high error detection rate and AdaBoost algorithm has good performance of on face detection,this study employs an improved algorithm that combine the advantages of the skin color segmentation and AdaBoost algorithm for face detection.Firstly,distinguish the face from skin color region using the skin color model,which can be used to initialize the AdaBoost cascade classifier for further getting rid of the non-face region from the candidate.The improved algorithm can improve the detection efficiency and reduce the detection error rate,through speeding up the detection speed using skin color information and optimized effect of detection.MATLAB is selected as the primary platform to test and simulate the algorithm.By testing the performance of the algorithm,it would be better to improve the detection and meet the requirement of experiment.Then the algorithm is transplanted into DSP hardware platform to perform simulation.The system is completed on ICETEK-DM6437-B Evaluation Module hardware platform.The core processor of the hardware platform is TM320DM6437.And then the algorithm and system functions are implemented in the CCS 3.3,which is DSP integrated development environment,and carried out the real-time debugging by its integrated real-time operating system DSP/BIOS.Optimize the detection algorithm,finally achieve the face detection.The experimental results show that the face detection system designed has a 90%accuracy rate,with a considerable robustness and practicality.This result is able to meet the basic real-time requirements.