The Performance Analysis Of Face Cascade Classifier

Face detection can be regarded as a specific case of object-class detection. The task is to find the locations and sizes of all faces in an image. Face detection has become an important research topic in the area of pattern recognition and computer vision, and been widely and increasingly used in many fields such as video retrieval, video surveillance and information security, etc.The thesis, taking face as detection object, studies the training technology of the classifier that can detect face quickly and accurately, and the performance analysis and application technology. First, the thesis describes the current face detection technology and the mainstream development. Second, the face detection method based on Adaboost algorithm and cascade algorithm proposed by Viola et al. is studied in details On this basis, an independent sample library meeting the training requirement is built. According to the thinking of the Adaboost algorithm, a strong classifier is constructed by extracting haar-like features from samples. Then the strong classifiers form a cascade classifier with cascade algorithm. Finally, a face detection system calls this classifier to achieve detection function. By analyzing the detection performance of classifier from the experiment data, the relationship between the sample library, training process and the detection performance is summed up.By comparing the detection performance with chart, the classifier training technology with better detection result can be achieved. Practice has proved that the face detection system using cascade classifier trained with such process can get the relative best detection performance, and proper detection rate, lower false detection rate. The topic, providing a good technical support for object detection application, has a strong versatility and portability.