Sensor Interoperability of Fingerprint Segmentation: An Empirical Study

Fingerprint segmentation is an important preprocessing step before feature extraction. Quality of fingerprint images acquired by various sensors of different types is distinctive. Fingerprint images from one data base have significantly distinct distribution from another in CMV feature space. The impact of sensor interoperability on fingerprint segmentation has received limited attention. This paper provides an empirical study on sensor interoperability of fingerprint segmentation. We find that a well trained fingerprint segmentation model on a single data set usually has higher accuracy on its homogenous testing data set, while lower accuracies on its heterogeneous testing data sets; and when a model trained on combined fingerprint data bases acquired from several sensors, the more training data sets to be combined, the more corresponding testing data sets achieve higher accuracies.