Active Sample Selection Algorithm And Its Application In Face Detection

Machine learning is an important branch of the artificial intelligence. Affiliated with machine learning, active sample learning selects representative samples from sample space actively to learn the distribution of data. It provides the theory and techniques for the complex pattern classification tasks including human face detection. To obtain the sample distribution with limited samples precisely, a hybrid cascade Bootstrap algorithm is proposed and applied to the face detection task. Based on the face detection, a method filtering objectionable information spread in internet and mobile telecommunication is also given in the paper.The main contributions of this paper are summarized as follows.To solve the sample selection problems in machine learning, an active hybrid cascade Bootstrap sample selection algorithm is proposed. Through formulation, theoretical comparison analysis indicates that the hybrid cascade Bootstrap algorithm, using a cascade multi-layer filter through iterations, can select more representative samples and handle the computation resource constraints on the size of training sample set. Thus a more effective classifier can be trained. Furthermore, both Bootstrap algorithm and hybrid cascade Bootstrap algorithm are applied to the negative sample selection in face detection based on AdaBoost algorithm. And the experimental results show the theoretical analysis is reasonable and the algorithm has good generalization capability.To handle the more and more severe overflow of junk information, an objectionable image detection method based on human face information is proposed. It extracts skin regions in images firstly, and then adopts physiology property of the ratio between human face and body to determine the objectionable level of images. Hybrid cascade Bootstrap algorithm is used to select representive samples for training the face classifier. The experimental results prove the effectivity of the method.