作　　者： ;

机构地区： 五邑大学智能技术与系统研究所

出　　处： 《中国图象图形学报》 2006年第1期19-25,共7页

摘　　要： 人脸的确认实质上是一个一类分类问题或野点检测问题,即只需要精确地描述某一类样本的分布,而将该类样本之外大范围内的样本点视为野点。为了能精确地描述某一类样本的分布,在对国内外现有统计学习理论和核方法进行研究的基础上,针对“人脸确认”这一特定的应用对象,分析了已有的一类分类算法,即支持向量数据描述方法在处理动态样本中存在的不足,进而指出,随着训练样本数目的增加,该算法会因为过大的优化规模而无法实际操作,为此提出了用于人脸确认的动态支持向量数据描述算法。由于新算法在优化过程中,仅需要考虑待检测样本和原有支持向量集,从而可以大大降低优化过程中涉及的运算规模和内存需求,进而可保证人脸确认过程中的实时性与动态性要求。 Face verification is essentially a prooblem of one-class classification or outlier detection. Its goal is to accurately describe one class of objects opposing to a wide range of other objects considered as outliers. Based on the existing research wook on statistical learning theory and kernel methods, this paper analyses the drawbacks of a existing one-class classification algorithm, namely support vector data description （SVDD）, on dealing with dynamic samples of face verification. This paper points out that, with the increase of samples, the size of optimization will exceed the memory space of the computer. Consequently, the algorithm will be unable to carry out. For the purpose of reducing the size of optimization, the dynamic support vector data description algorithm （DSVDD） is proposed. The new algorithm only computes the new samples and support vectors in the process of optimization, so that the required operation size and memory space can be reduced in a great degree, which means the real-time and dynamic demands are met for face verification.

关 键 词： 数据描述 人脸确认 核方法 野点

领　　域： [自动化与计算机技术] [自动化与计算机技术]