基于混合级联分类器的人眼检测器

Eye Detector Based on Mixed Cascade Classifier

在目标检测问题中,"级联"是一种常见的有效的思路。VJ的级联分类器存在着级分类器难以训练,复杂度不好控制等问题。Soft Cascade分类器则是一种复杂度可控的分类器,但是难以达到较高精度。综合了两者的优点,形成一种新的混合级联分类器。实验证明,混合级联分类器不但易于训练而且具有更好的精度。

When dealing with object detection tasks,Cascade is a common and effective idea.The VJ cascade classifier has disadvantages that some stage classifiers are difficult to train and the complexity is not easy to control.The Soft Cascade Classifier has lower complexity but lower accuracy at the same time.This paper combines the advantages of both so a new classifier called Mixed Cascade Classifier is proposed.