1 Biometric template selection and update: a case study in fingerprints Source:Pattern Recognition, Vol. 37, 2004, pp Authors: Umut Uludag, Arun Ross and Anil Jain Speaker: Shu-Fen Chiou( 邱淑芬 ) Date: 2004/11/04
2 Outline Introduction Proposed method Experimental results Conclusions Comment
3 Introduction A typical biometric system operates in two distinct stages: Enrollment stage. Authentication stage. In order to account for variations in the biometric data of a user, multiple templates corresponding to each user may be stored.
4 Problem partial overlap different sensorsa period of 6 weeks exhibiting difference in moisture content Result in inferior performance
5 Solve problems Store multiple templates Consider: number of templates storage computational Efficient function: automatic select templates
6 Proposed method Template selection DEND MDIST Template update BATCH-UPDATE AUGMENT-UPDATE
7 Template selection N fingerprint images Select K templates (K<N) Store templates in database Selection: Random selection Systematic selection Better performance
8 Template selection DEND
9 Template selection
10 Template selection MDIST Avg.
11 Template Update BATCH-UPDATE: newly acquired data. AUGMENT-UPDATE: after augmenting the new data set with current template set.
12 Experimental results DEND selection
13 Experimental results MDIST selection
14 Experimental results
15 Experimental results
16 Conclusion A systematic template selection procedure results in better performance than random template selection.
17 Comments 利用 DEND 的方法分完群後, 計算群裡面的 template 數量大小,加 上權重的觀念, 加快比對的效率。 經過一段時間測試,以權重而分的 template 被找到的次數再調整 template 的權重。 (a) weight=3 (b) weight=4 (c) weight=2 (d) weight=1 (e) weight=1
18 Comments 動態決定群的數量