非接触采集手形相对特征识别性能分析  

Recognition performance analysis for the hand-shape relative features acquired with contactless method

作　　者：苑玮琦[1] 荆澜涛[1] 吴微[1] 桑海峰[1] 

机构地区：[1]沈阳工业大学视觉检测技术研究所,沈阳110870

出　　处：《仪器仪表学报》2013年第9期2005-2012,共8页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(60972123);高等学校博士点专项基金(20092102110002);高等学校博士点专项基金(20122102120004);辽宁省教育厅科研项目(2009A561);沈阳市科技计划(F12-277-1-10);沈阳市科技计划(F10-213-1-00)资助

摘　　要：因非接触采集方式而产生的成像缩放与倾斜变形现象,使得成像后手形各特征的尺寸发生变化,从而造成绝对特征识别性能的显著下降。针对该问题作出了光学成像分析,证明了因倾斜变形现象产生的影响远小于缩放现象。因此在选用相对特征避免缩放现象的干扰后,利用类内方差均值和Fisher判别率,对变形后相对特征的稳定性和识别性能进行分析。实验结果表明,定义的相对特征可以作为稳定特征用于识别,同时经过小样本实验得到34个相对特征组成的向量,其识别率达到88.93%,说明该特征向量可以在小样本范围达到手形特征的有效识别,并可以作为手部识别的部分有效特征。Image scaling and inclination deformation phenomena generated with contactless acquisition make the hand-shape feature of the image change in size, which causes the recognition performance of the absolute features de- creasing markedly. In order to solve this problem, the paper makes optical imaging analysis. It is proved that the influ- ence of inclination deformation is much less than that of the scaling. Therefore, after using the relative features to avoid the scaling phenomenon, the mean of intra-class standard deviation （MSD） and Finsher＇ s discriminant ratio （FDR） are used to analyze the stability and recognition performance of the relative features after inclination deforma- tion. The experiment results show that the defined relative features can be used as stable features for recognition. At the same time, the recognition rate of the vector consisting of 34 relative features in small sample experiment is 88.93% , which proves that the feature vector can achieve the effective recognition of the hand shape in small sample space and can be used as the partial effective features of hand recognition.

关 键 词：非接触式采集 缩放现象 倾斜变形现象 手形相对特征 稳定性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]