Improving Face Detection Accuracy With SVM For Adaboost-based Algorithm

Face detection is one of the hottest research point of Pattern Recognition and computer vision research fields. But because of the influence of some factors, it is still a difficult work. Firstly, this paper introduces the research background and research meaning, the correlative knowledge and the pre-exiting primary detection methods, then makes a deep research and comparison on those advantage and disadvantage of those methods, and besides describes the face detection method based on Adaboost algorithm in detail.The face detection method based on Adaboost algorithm which is proposed by Viola and Jones is a popular detection method now, it makes a good behavior in detection speed and accuracy, but in the meanwhile it also has a high error-rate in some situation. On the basis of analysing the weekness of the method, this paper presents an approach to improve face detection accuracy with svm for adaboost-based algorithm. The priority and the difficulty of the method proposed is how to achieve a svm classifier with high accuracy and how to get the xml file of category feature library . Firstly, this paper uses the conversion of color space and a suitable way to fetch the characteristic value from the training samples, then train a svm classifier with high accuracy, and finally find out the human face areas from the candidate areas of the human face with the trained svm classifier. The experimental results show that svm algorithm can improve the detection accuracy and make the detection algorithm be a better detection efficiency.