Adaboost Algorithm Fast Object Detection Based On Integral Feature

Face detection is a fundamental and important research theme in the fields of pattern recognition and computer vision. A face is a normal and complicated nonrigid object, and an easily obtained information resouce in images and videos. Face detection, which is one of main parts in face expression understanding, not only has a great importance in theory, but has broad application value, such as face recognition, video monitoring, content-based retrival, ID authentication, video conference and so on. By employing integral type features, lip and eye detecton researches are done in this thesis. The main results in the thesis are as follows.Firstly, a lip integrated detection algorithm is proposed based on AdaBoost and Bayes discriminating features. A two-level cascaded AdaBoost classifier is construted using2-D Walsh features. After completely dumping negative training samples, target detected areas can be known by employing Bayes discriminating feature classfiers. The experiments show that the provided scheme has a better performance of false rate than Real AdaBoost on sef-established lip base.Secondly, a fast eye location scheme based on AdaBoost algorithms is proposed in view of detection speed and positioning accuracy. Fast and accurate eye location is one of key studies of face expressions. Eight extended feature forms are given combing Haar and triangular features. The experiments show that the scheme has a good detection performance, and improvement on positive recognition rate compared to that only using Haar features.Thirdly, some key problems on training samples are deeply analyzed. In this part, pre-processing of training samples, loading of negative samples, candidate area check by employing edge feature and skin color features, are discussed and studied, respectively. A median filter is given to eliminate noise and to remove the effect of illumination. The process of object detection and how to merge overlapping regions are introduced.