Research And Implementation Of Face Detection Based On AdaBoost Algorithm

With the rise of biometric recognition technology, face detection and recognition technology has become an important research topic of applied mathematics and pattern recognition. Face detection is a complex pattern classification problem and the first step for face recognition, and has a direct impact on the performance of the face recognition system. Face detection has a wide range of applications in video surveillance, content-based image retrieval and facial expression recognition system. Due to the complexity of face pattern, facial expression, gesture, light, acquisition conditions and other factors, how to establish a real-time, high detection rate face detection classifier is a great challenge and hot topic.The goal of this paper is to train a practical face detection classifier which can be embedded into face recognition system. The main work of this paper is described as follows:(1) Several skin detection methods in five color space were studied, a new skin detection method was proposed. The comparative experiment of several skin detection methods was conducted in the skin sample database, the proposed skin detection method in this paper has higher detection rate and low error rate was proved. The skin detection method was successfully applied to real-time video-based face detection, which improved the detection speed.(2) Face detection algorithm based on Haar features and AdaBoost algorithm was deeply studied, the difference of classification ability between single-threshold weak classifier and two-threshold weak classifier was analyzed. The experiment proved that the classification ability of two-threshold weak classifier is stronger than single-threshold weak classifier.(3) The traditional training method of AdaBoost algorithm is improved and the training speed is greatly increased. The pre-judgment mechanism is added to the test phase so that the detection speed can be improved. The trained visible light face detector and near infrared face detector was tested in the CAS-PEAL-R1database and CASIA-NIR database separately. The detection rate is98.64%and99.23%.In a word, the face detection module developed in this paper has a good, stable performance and successfully applied to the face recognition system developed by the Applied Mathematics Laboratory of Northeastern University.