Face recognition is a challenge in computer vision and artificial intelligence area,which exploits biological characteristics to identify the different persons. Due to itsadvantages, convenience and uniqueness, it has taken a wide applications in the areas ofinformation security, public security and so on.Face recognition system generally includes two aspects: face detection and facerecognition. Face detection is a process that detecting the human faces from static imagesor videos and extracting facial features. And face recognition is used to identify thedetected face from the Face Database.This paper analyzes the AdaBoost face detection algorithm and the EHMM facerecognition algorithm in detail. An AdaBoost based face detection algorithm using Haarfeatures is designed to detect the faces in the images or videos, which uses a22-stagecascaded classifier to decide whether a face is existed in the input image. Aslo anEnhanced Embedded Hidden Markov Model (EHMM) is designed to implement facerecognition. From top-to-bottom, the face image is divided into7super states by thedesigned EHMM to model the different face areas, including forehead, eyebrow, eyes, nose,upper lip, mouth and chin. For each super states, there are3or6embedded sub-states, sothere are total36embedded states used by the algorithm. Also this paper implements thedesigned face detection and face recognition algorithm on the Visual Studio platform, andlots of numeric experiments based on extended YaleB face database are performed. Theresults show that under the condition of normal illumination, the recognition accuracy canbe up to95%.Moreover, an embedded face recognition platform based on the TMS320DM3730isdesigned, which includes: Video acquisition module, Data sorage module, LCD module,Data transfer module, Man-machine interactive module, Power management module andso on. Finally, Fixed point algorithm for face detection and face recognition are designedand transplanted to the designed DM3730platform. Testing results show that about60M cycles are needed to implement face detection in an image range of320*240pixels, andabout30M cycles are needed to match the detected face with one face image in thedatabases, which meets the real-time application well. |