Image versus feature mosaicing: A case study in ﬁngerprints Arun Ross, Samir Shah and Jidnya Shah West Virginia University, Morgantown, WV, USA. ABSTRACT Fingerprint mosaicing entails the reconciliation of information presented by two or more impressions of a ﬁnger in order to generate composite information. It can be accomplished by blending these impressions into a single mosaic, or by integrating the feature sets (viz., minutiae information) pertaining to these impressions. In this work, we use Thin-plate Splines (TPS) to model the relative transformation between two impressions of a ﬁnger thereby accounting for the non-linear distortion present between them. The estimated deformation is used (a) to register the two images and blend them into a single entity before extracting minutiae from the resulting mosaic (image mosaicing); and (b) to register the minutiae point sets corresponding to the two images and integrate them into a single master minutiae set (feature mosaicing). Experiments conducted on the FVC 2002 DB1 database indicate that both mosaicing schemes result in improved matching performance although feature mosaicing is observed to outperform image mosaicing. Keywords: Image mosaicing; Feature mosaicing; Sum rule; Thin-plate splines (TPS); Elasticity; Correlation. 1. INTRODUCTION Compact ﬁngerprint sensors can now be easily integrated into devices such as laptops, cellular phones, PDAs, etc. The limited platen size of these sensors results in the acquisition of ﬂat/dab prints that contain reduced information (e.g., fewer minutiae points) compared to rolled ﬁngerprints 1, 2 as seen in Figure 1. Thus, multiple impressions of the same ﬁnger may have only a small region of overlap, thereby degrading the matching per- formance of the ﬁngerprint authentication system. In order to address this problem, information from several impressions of a ﬁnger can be integrated to enhance the information content of the resulting ﬁngerprint template. This process, known as mosaicing, is expected to improve the matching performance of a ﬁngerprint system. Mosaicing can be accomplished at two distinct levels: (a) the image level (image mosaicing), where multiple impressions of a ﬁnger are combined together to generate an elaborate ﬁngerprint - minutiae points are then extracted from the mosaiced ﬁngerprint; and (b) the feature level (feature mosaicing), where the minutiae sets extracted from multiple impressions are combined to generate a composite feature set (Figure 2(a) and (b)). Both these schemes rely on a robust registration technique to accurately align a pair of impressions (or minutiae sets) before integrating them. One of the confounding factors in ﬁngerprint registration is the presence of elastic deformation in constituent prints. The elastic deformation is a consequence of the elasticity of the skin and the non-uniform pressure applied by the ﬁnger on the platen of the sensor. This leads to a distortion of the ridges and the accompanying minutiae points. Thus, a simple aﬃne transformation is not suﬃcient to register two such ﬁngerprint impressions (or minutiae sets). The goal of this paper, in the context of mosaicing, is two-fold: 1. To incorporate a registration procedure that compensates for the non-linear distortion present between a pair of prints. 2. To compare the utility of a composite template generated by image mosaicing with that of feature mosaicing by examining their matching performance against a target set of prints. Further author information: : (Send correspondence to Samir Shah) Arun Ross: E-mail: Arun.Ross@mail.wvu.edu, Telephone: 1 304 293 0405 ext 2556 Samir Shah: E-mail: sshah@csee.wvu.edu, Telephone: 1 304 293 0405 ext 2539 Jidnya Shah: E-mail: jidnyas@csee.wvu.edu, Telephone: 1 304 293 0405 ext 2539 Proc. of SPIE Conference on Biometric Technology for Human Identification III, (Orlando, USA), pp. 620208-1 - 620208-12, April 2006.