基于独立成份分析和支持向量机的人眼注视研究

Investigation of Eye Gaze Based on Independent Component Analysis and Support Vector Machine

采用独立成份分析(ICA)方法对单眼宽约为40 pixel的低分辨率人眼图像进行预处理,并把得到的一组组合系数输入支持向量机(SVM)判断图像中人眼的注视与否。对于单用户的注视探测,一维组合系数输入实现了100%的正确识别率;对于多用户的注视探测,二维组合系数输入实现了100%的正确率识别率。

Different from the costly gazhing direction measurement and gazing tracking systems, the gazing detection system always uses low-resolution images to realize human-machine interaction. In this paper, the independent component analysis （ICA） is applied to preprocess the low-resolution （40-pixel wide of a single eye） eye images. After the processing,each eye image,which is represented by a low-dimensional vector,will serve as the input of Support Vector Machine（SVM） to give the judgment of gazing or not. The 1-dimensional and 2-dimensional vector input have achieved 100M classification rate to single-user and multi-user, respectively.