The Design And Implementation Of The Face Online-detection System Based On RLAB Feature

The primary mission of face detection is to judge whether there are faces in image. If so, further find out face information such as the location and scope. Face detection was initially proposed as a positioned part of face recognition system, which was mainly used for the pretreatment of face recognition. In recent years, face detection gradually becomes a new research focus, widely used in video, intelligent robotics and medicine. Such as laptops and mobile phones provide login identification, album category and smile face detection photograph and many other convenient applications. Face is a biometric which has a friendly, direct and obvious characteristic. But the face of great randomness local features and resulting in changes over light, shelter and other factors, so that the the face detection becomes a very challenging subject. Domestic and foreign researchers have proposed many effective face detection algorithms. They can be divided into knowledge-based approach, invariant feature Method, template matching method and method based on statistical theory.In this paper, the face online-detection system is based on statistical learning method which is popular and more efficient. It consist of communication module, face detection and classifier training. The communication module is responsible for receiving client requests recognizable image and returning the result. It is a C/S architecture based on I/O completion port，the best communication model IOCP of Windows, and achieves a high throughput and low latency. The face detection module uses a cascade classifier to search faces at different locations and scales by reducing the image on the feature oriented and wnidow oriented ways. Model training produces a high-performance cascade classifier through image normalization, feature space calculations and weak classifier selection, strong classifier construction etc.Face detection is the front face recognition system. Its results have a direct impact on the realization of subsequent location tracking and recognition. To improve the performance of the algorithm, this paper presented a new feature RLAB (Random Locally Assembled Binary), based on the traditional Haar feature and LAB (Locally Assembled Binary) feature, which is used in classifier training and face detection process. Due to the use of strategies and advanced methods, such as integral image, lookup table type weak classifiers, cascade classifier, the face online-detection system have achieved good results even if detect face images from a different size, angle, background, illumination, expression and other aspects. It shows the effectiveness of the algorithm, and wide application value of the face online-detection system.