Face Detection And Recognition

With the rapid development of modern technology, the degree of living intelligenceis increasing constantly. Many kinds of intelligent systems are applied in many fields. Inthese systems, intelligent identification system is paid more attention because of itswidespread application, especially based on face recognition. Compared with thetraditional system based on the fingerprint and iris recognition, face recognition is moreconvenient and user-friendly. Because of these advantages, face recognition technologyhas drawn more and more attention.In the recent years, depending on the more investment in the research of facerecognition, the technology gets rapid development. Many company and researchinstitutions have put forward face recognition systems. The generation of these systemsis poor, because it is sensitive to illumination variation and poses variation. Most ofthese algorithms overcome either of the variations, but fail to both.Based on the above prospect and limitation, the paper investigates face recognitiontechnology with various factors. We focus on video from monitoring system，it has highreal-time requirement. Therefore, the proposed method not only overcome the influencecaused by illumination variation and poses variation, but also improves the recognitionrate. The contribution of this paper is arranged as the following:The framework of face detection is based on skin model and Adaboostalgorithm.The size of image is720×560, so the real-time requirement is difficult withAdaboost detection algorithm directly. For this reason,the paper designs a multi-stageface detection method. We firstly use skin model for pre-detection in order to getprobable regions of face in the image Then we we accurately detect face in thoseprobable regions with Adaboost algorithm. The experimental results show that thecombination of skin model and Adaboost algorithm make face detection more rapid andeffective.Face recognition based on EHMM under difficult lighting condition: In order toimprove the accuracy of face recognition under various illumination and poses, wepropose a method of face recognition based on combination of Local Binary Pattern (LBP) and embedded hidden Markov Model (EHMM).In order to overcome posevariation, we firstly divide the object into many patches in the image. Then, we useEHMM to describes object with patch information locally and the relationship betweendifferent patches globally. It describes the human face in different pose effectively. Forillumination variation, image which is transformed into feature map by LBP operator isnormalized under various lighting conditions firstly; Then, the feature is extracted byDiscrete Cosine Transform (DCT) in feature map.The feature extracted is used for patchinformation in EHMM. The experimental results show that the combination of them canimprove the rate of face recognition in complex scenes.