Research On Eigen - Feature Extraction Based On Joint Haar Feature In E - Learning

The research of this paper is funded by National Natural Science Foundation of China project "The analysis of emotion recognition based on academic expression in E-Learning"(NO.60970052) and Beijing Natural Science Foundation project "The Study of Personalized E-Learning Community Education based on Emotion Psychology"(NO.4112014).The research of this paper is to extract expression characteristic parameters of the learner in E-Learning, which provides necessary parameters for emotional modeling and emotional teaching in the subsequent research. The research of this paper consists of three parts:firstly, we can capture learner’s facial image by the camera installed on the computer; Then, we can detect and locate learner’s face image using image processing teciniques; Finally, we can locate the eyes and mouth area from the face image, and extract the expression characteristic parameters of the learner. This paper selects four important parameters to extract, which concludes:the face area, the distance of eyelids, up mouth radian and down mouth radian. The main-work and achievements of this paper are as follows:(1) Due to the real-time requirement of E-Learning system, this paper proposes a decile eigenvalue Adaboost algorithm and joins FPR(False Positive Rate) in this algorithm. Experiments used this improved Adaboost algorithm show good speed and accuracy performance when there’s no skin-color thing in front of face.(2) Traditional face detection algorithm based on Haar features have lower detection rate when there’s skin-color thing in front of face, this paper provides a new method combines co-occurring Haar features and the improved Adaboost algorithm in(1) to detect and locate face area. Experiments use this new method show good performance when there are some skin-color things in front of face, and it also have good performance in detection rate and speed.(3) Due to the expression characteristic parameter requirement of E-Learning system, we use maximum variance between-class segmentation, intergral projection, regionan center positioning and Harris corner detection method to extract the distance of eyelids and mouth radian. Experiments achive good performance.(4) Based on the theoretical methods for the above, we design and implementation anexpression extraction system in E-Learning environment. Experiments show that this system can meet the real-time requirement of E-Learning, it also shows good performance in the accuracy of expression parameters it extracts, which provides parameters for the subsequent emotional modeling research.