Hsun-Li Chang｜國立臺北商業大學資訊管理系 助理教授｜hsunli@ntub.edu.tw
Kai-Yung Lin｜育達科技大學資訊管理系暨研究所 副教授｜linky@ydu.edu.tw
Chih-Chiang Yu｜育達科技大學資訊管理系暨研究所｜ta0614@hotmail.com
▌Abstract
Attention is closely related to learning. How to measure students’ attention levels is a very important topic. Among a lot of measuring approaches, face detection is more objective and has less effect on students. In this paper, we proposed a novel face detection approach based on the Extended Haar Texture Features (EHTF) and the EHTF-Boosting learning algorithm. EHTF uses 6 differential rectangular feature templates to describe characteristics of edge, linearity, orientation, and spot in the images’ local area. EHTF has benefits to evaluation faster and illumination invariant. Applying EHTF on the training face/non-face images, EHTF-Boosting is used to train several weak classifiers to form a strong classifier. The proposed EHTF-Boosting learning algorithm can correctly show the importance, i.e. the weight value, of each weak classifier. By using 2,000 frontal face images and 2,000 background images from LFW-Bigfoto dataset as the training set, we implemented a face detector based on the proposed methods and applied it on the MIT-CMU testing dataset. Comparing EHTF with other methods, the experimental results show that our approach can improve detection rate and reduce the false positive rate obviously. It has several advantages: using fewer features, less training time and higher detection rate. The proposed method can also satisfy the requirements for real-time applications and can be applied on different instructional environments as the data acquiring method, for example online learning, attendance analysis, learning difficulty analysis, and keeping order in course. Analysis data acquiring for related researches can be more convenient and objective.
Keywords: Attention, Face Detection, AdaBoost, EHTF-Boosting