Adaboost人脸检测算法的改进

AdaBoost人脸检测算法是该领域中比较成功的算法之一。当被检测图片中人脸大小缩放至与分类器尺寸左右时,才能够正确地判定人脸,而其他缩放下的检测造成了冗余。验证了人脸检测速度的影响因素,针对分辨率为640×480的图片,建立了人脸尺寸高斯模型以及单幅人脸权重模型,提出了依据这两个模型的人脸检测的缩放方式。实验显示基于该方法,在稍有降低识别率的情况下,大幅提高了检测速度。

The Improvement of Adaboost Face Detection Algorithm

Adaboost face detection algorithm is one of the most successful algorithms in this field. When a picture is scaled until the size of human face in it is near the size of the classifier, we are able to determine the correct face. But the detection under other scales resulted in redundant. In this article, we verify the factors which influence the detection speed. A face size Gaussian model and weight of single human face model are built for the images whose resolutions are all 640x480. Then the scaling of face detection is proposed according to these two models. Experience has shown that this model can reach a high substantial increase in detection speed under in the case of a slight decrease in the recognition rate.