Research Of Ada Boost-based Face Detection Algorithmn

Machine vision is an important part of science of artificial intelligence. Face analyzing is one of the difficult technology in Machine vision, which includes face detection and face recognition. Face Detection which is the key to intelligent recognition has important research value. According to the historical development process, face detection from the initial focus on improving the system accuracy, to ensure the accuracy of the present system under the premise of improving the speed and achieve real-time during the course of a large number of efficient detection algorithms have been proposed and developed. The most representative one is AdaBoost-based face detection algorithm. The face detection technology by integrating map and classifier cascading and other methods can effectively reduce system detection time and improve system efficiency, and ensure a high level of accuracy, thereby achieving a real-time system standards. This method was later improved by many scholars to study, currently in testing the overall performance of the integrated performance is very good. It is a learning of the sample-based detection methods which is characteristics-based. But the training speed is less considered.Because of the fast update of modern database, the importance of training efficiency increased. This paper will enhance the training efficiency and to focus on research as a breakthrough. In AdaBoost algorithm, First, by PAC learning model, Haar-like features, integration plans, training and detection classifier cascade of five key technologies, described by the concept and theory derivation algorithm is discussed in detail the nature of each stage, as the theoretical basis discussed later. Then experiment by training cost analysis illustrates the need for feature selection; eigenvalue analysis by feature selection experiment illustrates the effectiveness and feasibility of the characteristics of the merits of the measure.With experimental results, it begins to introduce the proposed feature selection method in detail. A preprocessing filter method has been introduced first. The method is based on evaluation theory, evaluation obtained by experimental analysis, the feature can be selected in proportion. This filtration process to ensure that all of the features out of heterogeneous samples tested for differences in the characteristics of less than all reservations. Then introduces a method for filtration with the cascade synchronization. The method uses cascading structure from simple to complex features, relying on the relative contribution of features for feature selection. In the sample database update process, this method has good adaptability. In this paper, two methods can be integrated based on actual use. Finally, feature selection experiments performed to verify the characteristics of the relevant parameters on the properties, the paper describes the characteristics of filtration methods for theoretical and practical significance, and prospects for future research.