A Study Of Face Detection Based On Adaboost Algorithm

Face detection is a necessary first-step in face recognition systems. Given an arbitrary image, the goal of face detection is to determine whether or not there are any faces in the image and, if present, return the image location and extent of each face. Because of its great application value in surveillance systems, content-based linage detection systems, intelligent human-computer interfaces and soon, it has been approved to bean independent subject and received wide attention from a great many of researchers in the se year. In this thesis, the Adaboost cascade face detection algorithm proposed by Viola et a1.is analyzed in detail. The main contributions are as follows:1. A great amount of literatures, surveys and research papers concerning up-to-date techniques of face detection and face recognition are read and analyzed. Some hot issues about face detection are discussed and studied in this paper.2. Haar feature, integra1 image, weak classifier, strong classifier are introduced in detail in the thesis, on the other hand, a face detection system is built while in-depth study on face detection method based on Adaboost algorithm.3. Focusing on the disadvantages of Adaboost algorithms: When the detector encounters faces rotated, there always occurs center undetected area. So we propose two new feature templates, which improved the detect effect greatly. And in order to optimize the detector, the human skin model and illumination compensation are added in front of the detector.4. The training of the adaboost algorithm always spends a lot of time, to resolve the problem, the thesis analyzes the feasibility of paralleling the training algorithm. Then the parallelized algorithm is introduced. In the end, experiments showed that, the parallelized algorithm reduced the training time by 2 / 3 or more.