,1190 北京科技大学学报 2006年第12期 [2]Hurley DJ.Nixon M S,Carter J N.Force field energy fune- [6]Hyvarinen A.Oja E.A fast fixed-point algorithm for inde- tionals for image feature extraction.Image Vision Comput. pendent component analysis.Neural Comput,1997.9(7): 2002,20(5):311 1483 [3]Chang K.Bowyer K W,Sarkar S,et al.Comparison and [7]Carreira-Perpinan M A.Compression neural networks for fea- combination of ear and face images in appearance-based biomet- ture extraction:application to human recognition from ear im- rics.IEEE Trans Pattern Anal Mach Intell.2003.25(9): ages [Dissertation]Spain:Technical University of Madrid. 1160 1995 [4]王忠礼，穆志纯，王修岩，等.基于不变矩匹配的人耳识别· [8]Hyvarinen A.Fast and robust fixed point algorithms for inde- 模式识别与人工智能，2004,17(4)：502 pendent component analysis.IEEE Trans Neural Networks [5]Bartlett M S.Face Image Analysis by Unsupervised Learning 1999,10(3):626 and Redundancy Reduction [Dissertation].University of Cali- [9]Vapnik V N.统计学习论理的本质.张学工，译.北京：清 fornia San Diego.1998 华大学出版社，2000 Ear recognition based on compound structure classifier ZHA NG Haijun,MU Zhichun) 1)Information Engineering School.University of Science and Technology Beijing.Beijing 100083.China 2)Department of automation,Shenyang Institute of Aeronautical Engineering.Shenyang 110136,China ABSTRACI Based on the research of ear recognition with independent component analysis(ICA),a new compound structure classifier(CSCER)ear recognition model was proposed.The model made rough classi- fication to the human ears first according to their geometric features,then ICA was used to extract the alge- bra features and support vector machine (SV M)was for detailed classification,finally the results were achieved,which was in accordance with human natural recognition process.The model overcame the single ICA disadvantages of costing too much time and with too many features,also avoided losing structure fea- ture when ear images were preprocessed.The experiment shows that the model can achieve high recognition rate and is suitable for complex ear image libraries. KEY WORDS ear recognition;independent component analysis;SVM;structure classifier［2］ Hurley D J‚Nixon M S‚Carter J N．Force field energy func￾tionals for image feature extraction．Image Vision Comput‚ 2002‚20（5）：311 ［3］ Chang K‚Bowyer K W‚Sarkar S‚et al．Comparison and combination of ear and face images in appearance-based biomet￾rics．IEEE Trans Pattern Anal Mach Intell‚2003‚25（9）： 1160 ［4］ 王忠礼‚穆志纯‚王修岩‚等．基于不变矩匹配的人耳识别． 模式识别与人工智能‚2004‚17（4）：502 ［5］ Bartlett M S．Face Image Analysis by Unsupervised Learning and Redundancy Reduction ［Dissertation］．University of Cali￾fornia San Diego‚1998 ［6］ Hyvarinen A‚Oja E．A fast fixed—point algorithm for inde￾pendent component analysis．Neural Comput‚1997‚9（7）： 1483 ［7］ Carreira-Perpinan M A．Compression neural networks for fea￾ture extraction：application to human recognition from ear im￾ages ［Dissertation ］．Spain：Technical University of Madrid‚ 1995 ［8］ Hyvarinen A．Fast and robust fixed-point algorithms for inde￾pendent component analysis．IEEE Trans Neural Networks‚ 1999‚10（3）：626 ［9］ Vapnik V N．统计学习论理的本质．张学工‚译．北京：清 华大学出版社‚2000 Ear recognition based on compound structure classifier ZHA NG Haijun 1‚2）‚MU Zhichun 1） 1） Information Engineering School‚University of Science and Technology Beijing‚Beijing100083‚China 2） Department of automation‚Shenyang Institute of Aeronautical Engineering‚Shenyang110136‚China ABSTRACT Based on the research of ear recognition with independent component analysis （ICA）‚a new compound structure classifier （CSCER） ear recognition model was proposed．The model made rough classi￾fication to the human ears first according to their geometric features‚then ICA was used to extract the alge￾bra features and support vector machine （SVM） was for detailed classification‚finally the results were achieved‚which was in accordance with human natural recognition process．The model overcame the single ICA disadvantages of costing too much time and with too many features‚also avoided losing structure fea￾ture when ear images were preprocessed．The experiment shows that the model can achieve high recognition rate and is suitable for complex ear image libraries． KEY WORDS ear recognition；independent component analysis；SVM；structure classifier ·1190· 北 京 科 技 大 学 学 报 2006年第12期