基于主成分分析和支持向量机的人眼注视识别

Recognition of eye gazing based on principle component analysis and support vector machine

眼睛注视是人类传播信息的一个重要的媒介,实现对人眼注视与否的探知已经成为人机交互领域中一个亟需解决的问题。针对这个问题,首先采用主成分分析方法对人眼图像进行预处理,得到有利于人眼分类的一组低维特征系数,然后采用基于统计学习理论的支持向量机网络实现对人眼注视与否的判断,并通过大量的实验得出适合于人眼注视与否的最佳主成分特征维数。实验表明,在有限个样本的训练下,当PCA特征维数在18至23之间时,能够得到令人满意的结果,正确识别率均在91%之上,而特征维数为20时,测试样本被正确识别的比率最高,可达98.78%。

Eye gazing is an important medium of human communication, achieving the judgment of eyegazing with computers has become an urgent problem to be solved in the field of human-computerinteraction. For this problem, firstly this paper uses the principle component analysis to reprocess theeyes＇ images to get a set of low-dimensional characteristic coefficients, which is useful for the human eyeclassification, then the support vector machine which is based on statistical learning theory is used to giveout the judgment of gazing or not, and through a large of experiments, the feature dimension range whichare best suitable for the problem can be obtained. The experimental results show that when using limitedtraining samples, and the dimension of the principle component analysis characteristic coefficients isbetween 18 and 23, it can get a satisfactory result with its correct ratio more than 91%, especially whenthe feature dimension is 20, the proportion that the testing samples are identified correctly can beobtained as high as 98.78%.