The Research And Implementation Of Mutli-view Faces Detection System

After several decades of development, face detection technology has achieved encouraging results, especially in the frontal face detection, the current algorithm can already achieve high detection rate and quick speed. However, because of the diversity of multi-view faces, there are still many deficiencies in the research of multi-view detection. This paper describes a multi-view faces detection system, which have higher detection rate and faster speed.Viola's face detection algorithm is based on Adaboost algorithm. The algorithm has been proven to be an effective face detection algorithm. This paper designs a multi-view faces detection system based on the viola's face detection algorithm. The system includes the following three parts: frontal faces detection, profile faces detection and non-upright faces detection. They were designed corresponding algorithms. On the frontal faces detection algorithm, in order to improve the speed of training classifier, we will have direct access to the objective function, avoid each classifier was calculated repeatedly, eventually weak classifier training process turn into a process of cumulative weights; on the profile faces detection, in order to make classifier have a high detection rate, this paper expanded Harr-Like features, and used these new features in the process of training; on non-upright faces detection, in order to avoid training new classification every time, this paper use rotating detection window method. To maintain the gray features, this paper used Gaussian template rotating algorithm to rotate detection windows; during the process of combined these algorithms, in order to accelerate the speed of system, we revised the original detection algorithm to eliminate redundant testing, and to accelerate the mobile speed of the detection window.Numerous experiments have proved that the detection system has higher detecti- on rate and speed.