Then a two-level method is proposed to evaluate the sensor interoperability of commonly used features. This method preliminarily Cisplatin Sigma evaluates the feature(s) through the first level evaluation based on segmentation error rate, the feature(s) whose segmentation error rate is high and not stable will be eliminated; the remaining candidate feature(s) will participate in a second level evaluation Inhibitors,Modulators,Libraries which is based on a decision tree, and the feature or feature set with good sensor interoperability will be selected according to information theory. The effectiveness of the proposed method is validated by experiments performed on a number of fingerprint databases derived from various sensors.The paper is organized as follows: Section 2 analyzes the sensor inoperability problem of features in fingerprint segmentation through empirical studies.
Section 3 proposes our two-level feature evaluation method. Section 4 reports the experimental results. Finally, Section 5 draws conclusions and discusses future work.2.?Sensor Interoperability Problem of Feature in Fingerprint SegmentationIn fingerprint segmentation, features are an important topic and discriminating features Inhibitors,Modulators,Libraries usually leads to favorable segmentation performance. There is abundant research on segmentation features, which mainly focuses on defining the discriminating features. The commonly used features in fingerprint segmentation include gray-level features [2,3,9,17,18] (such as gray mean, gray variance, contrast, etc.), texture features [3,5,7,9,19,20] (such as gradient, coherence, Gabor response, etc.
), and other features [12,21�C23] (such as Harris corner point features, polarimetric feature, number of invalidated minutiae, etc.).Due to the differences between Inhibitors,Modulators,Libraries sensing technologies, fingerprint images derived from different sensors usually have different characteristics, resolution, quality and so forth. Therefore, various features have different discriminating abilities on images derived Inhibitors,Modulators,Libraries from different sensors. In order to investigate the influence of various sensors on the segmentation feature, we randomly selected fingerprints from a number of open databases and analyzed the feature histograms.The fingerprints were collected from three open fingerprint databases: FVC2000 [24], FVC2002 [25], and FVC2004 [26]. Each open database contains four sub-databases, where the first three sub-databases are derived from three different types of sensors, and the last sub-database is generated synthetically.
The sensors used in the open databases are presented in Table 1. Each database consists Cilengitide of a training set of 80 images and a test set of 800 images.Table 1.FVC fingerprint database sensor list.We randomly select 10 fingerprint images from each real sub-database to construct a database containing 90 images. Each of the 90 images is partitioned into non-overlapping blocks of 8 �� 8 pixels, and then all the blocks are manually www.selleckchem.com/products/Bosutinib.html labeled as the foreground class and the background class.