摘要
首先开发了特征脸识别系统,然后采用自制的、MIT的和普渡大学的人脸数据库,用实验研究了将较多的人脸样本模式部分相交分组或不相交分组时,无拒和有拒情况下所开发识别系统的识别率。实验研究表明:在互不相交的分组中,单个样本训练的识别率远低于多样本训练的情况,识别的结果依赖于训练样本和样本数据库;在部分相交的分组中,无拒和有拒情况下的识别率均优于无相交时的识别率,这主要是由于部分相交的分组增加了识别系统的容错性。
Based on the eigenface method, a face recognition system is first developed. And then the system recognition ratios are investigated when the sample face patterns are partly intersected or not intersected by utilizing the self made,the MIT and the Purdue University face databases respectively. Experimental results indicate that the recognition rate of single training sample is much lower than those of many ones in non intersection groups; the recognition rates depend on the given databases; and the recognition rates under the conditions with rejected or non rejected samples are higher in small intersection groups than those in non intersection ones. It is because the style of intersection groups improves the fault tolerance of the system.
出处
《计算机工程与科学》
CSCD
2005年第6期38-39,42,共3页
Computer Engineering & Science
关键词
识别率
相交分组
人脸数据库
identification rate
intersection group
face database
