Adaboost人脸检测方法及其并行实现  

Face Detection Based on Adaboost and Parallel Realization

作　　者：左登宇[1] 董兰芳[1] 宋波[1] 

机构地区：[1]中国科学技术大学计算机系,安徽合肥230027

出　　处：《计算机仿真》2010年第6期249-253,257,共6页Computer Simulation

摘　　要：人脸检测是计算机视觉领域的基础研究,在视频监控、自动人脸识别等领域有着重要应用价值。根据Adaboost方法的人脸检测算法因快速和检测率高,得到了成功的应用。但是方法训练出的检测器对倾斜的人脸存在检测盲区,同时Adaboost算法训练的计算复杂度高,需要存取的数据量非常大,导致训练过程显得非常缓慢。针对上述问题,为了提高检测精度和速度,提出了两种新的特征模板加入到训练过程中,改进了其对倾斜人脸的检测效果。并给出了训练过程的并行算法,大大减少了训练时间,为解决训练耗时问题提供了一个新的途径。Face detection is a fundamental research theme in the topic of Computer Vision, and it has a broad application in many fields such as video surveillance, automatic face recognition, etc. In recent years the face detection algorithm, which based on Adaboost, has been successfully applied because of its rapid speed and acceptable detection rate. However, there are certain blind spots on tilted face detection. On the other hand, the Adaboost learning algorithm has a high computation load and data throughput, as a result, the training process takes a lot of time. To deal with these problems, two new Haar features are proposed in the training process, which have improved the detection effects on tilted human faces. A parallel Adaboost algorithm has also been proposed in this article. It reduces the training time greatly and provides a new path to solve the time consuming problem in the training process.

关 键 词：训练 特征 弱分类器 并行化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]