Analysis and improvement of face detection based on surf cascade

2nd Annual International Conference on Information System and Artificial Intelligence

物理学;计算机科学

Hu, Siquan^1 ; Zhang, Caihong^1 ; Liu, Lei^1

School of Computer and Communication Engineering, University of Science and Technology, Beijing

100083, China^1

关键词: Cross validation;    False positive;    Frontal faces;    Keypoints;    Logistic regressions;    Weak classifiers;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/887/1/012027/pdf
DOI  :  10.1088/1742-6596/887/1/012027

学科分类：计算机科学（综合）

来源: IOP

【 摘 要 】

This paper aims to study limitations of the commonly employed boosting cascade framework. We focus on the factors like data, feature, weak classifier and stages. A set of novel experiments were done to show the relationship. The model contains three key points: SURF feature, weak classifier based on logistic regression and AUC-based cascade learning algorithm. This paper adds cross validation in logistic regression creatively which improves accuracy and speeds up convergence greatly. Eventually only five stages and about 100 weak classifiers are needed. The frontal face detector improves reject rate to 99% for the first three stages, decreases number of false positive greatly and achieves comparable performance among non-CNN techniques on FDDB dataset.