A Linear Hybrid Classifier for Fingerprint Segmentation

Fingerprint segmentation is the important step of image preprocessing in an automatic fingerprint identification system and usually aimed to exclude background regions to reduce the time of subsequent processing and avoid detecting false features. In this paper, a hybrid algorithm based on linear classifiers for the segmentation of fingerprints is presented. The propose algorithm uses a block-wise classifier to separate foreground and background blocks in the main, and employ a pixel-wise classifier to deal with pixels accurately. In order to evaluate the performance of the new method in comparison to the methods based on other classifiers, experiments are performed on FVC2000 DB2. The average error rate of the hybrid technique is observed to be 0.53%, while that of the label box-based segmentation is 0.80%.