摘要
提出了一种基于多示例的Boosting级联算法,通过使用多示例、大的训练集以及对应的阈值调整方法,提高了人脸检测速度和精度。实验证明,该方法在进行人脸检测时达到了非常高的检测率,并且速度比其它算法提高了1￣2倍。
This paper illustrates a method for real-time face detection, which uses a large scale training set, an evaluation set based on multiple instances, and a cascade classifier based on Boosting, called Multiple Instance Boosting Cascade. With divided training sets as an offline set and an online set, detector＇s performance is improved by bootstrapping samples during training on large scale data. Under the utilization of various parameters to adjust the strategies of discarding evaluation data, and threshold adjustment based on multiple instance samples, final results surge in outstanding face detection accuracy and speed. Compared with other detectors, this method achieves better ROC curves and 1-2 times faster speeds.
出处
《软件导刊》
2008年第5期98-99,共2页
Software Guide
