CAJ | 학술논문

针对在已有人脸检测方法中采用单阈值所导致的误检率太高的问题，提出一种基于优化加权参数的快速Ad-aBoost训练检测算法。算法通过改变弱分类器加权参数求解公式的方法，保证了在低误检率的前提下也能获得低误警率；通过特征值曲线自适应得到双阈值，然后构造双阈值弱分类器并进行集成，形成强分类器。实验结果表明，该算法不仅能够有效地提高检测精度，而且，由于双阈值能够减少搜索次数，从而使训练和检测时间也有明显的改进。
침대재이유인검검측방법중채용단역치소도치적오검솔태고적문제，제출일충기우우화가권삼수적쾌속Ad-aBoost훈련검측산법。산법통과개변약분류기가권삼수구해공식적방법，보증료재저오검솔적전제하야능획득저오경솔；통과특정치곡선자괄응득도쌍역치，연후구조쌍역치약분류기병진행집성，형성강분류기。실험결과표명，해산법불부능구유효지제고검측정도，이차，유우쌍역치능구감소수색차수，종이사훈련화검측시간야유명현적개진。
Existed single threshold methods will cause highly false recognition rates in face detection. This paper proposes a fast AdaBoost algorithm based on optimized weighting parameter. Firstly, algorithm changes the solving formula of weighted parameter to get low false alarm rates even under the premise of low false recognition rates. Secondly, dual-threshold is obtained by calculating the feature-value curve. Finally, the detected dual thresholds are used to form weak classifiers, which can form an ensemble classifier. Experimental results show that it not only improves the accuracy of detection, but also ameliorates the training and detecting time since dual-threshold can decrease the times of searching.