A gradient-based combined method for the computation of fingerprints’ orientation field
Introduction
Accurate automatic personal identification is critical in a wide range of application domains such as smartcard, electronic commerce, and automated banking. Biometrics, which refers to identifying an individual based on his or her physiological or behavioral characteristics, is inherently more reliable and more capable in differentiating between an authorized person and a fraudulent imposter than traditional methods such as knowledge-based [password or personal identification number (PIN)] and token-based [passport or driver license]. Among all biometric traits, fingerprints have one of the highest levels of reliability [1] and have been extensively used by forensic experts in criminal investigations [2], so designing an automatic fingerprint identification system (AFIS) with high accuracy has very important significance.
Although automatic fingerprint recognition has been extensively studied and has received good performance on small database, there still exist some critical issues such as long processing time on large databases and low matching rate on poor image. To solve these problems, improvements on fingerprint classification and identification are needed. As a global feature of fingerprint, orientation field which describes the local direction of the ridge-valley pattern, plays a very important role in both topics mentioned above.
During the past years, lots of methods have been proposed for calculating fingerprints’ orientation fields, which can be broadly categorized as gradient-based approaches [3], [4], [5], [6], filter-bank based approaches [7], [8], and model-based approaches [9], [10], [11], [12], [13]. Filter-bank based methods are resistant to noise, but their results are not very accurate because of the limited number of filters, furthermore, they are also known to be computationally expensive due to the comparison of all filters’ outputs. Model-based methods try to consider the global constrains and regularities of orientation fields except for the areas around singular points [12], so they are able to predict orientation fields for the large noise areas, but almost all model-based methods depend on accurate extraction of singular points, and for the poor fingerprint images, it is a hard work. At the same time, model-based methods often can not give out accurate orientation fields for the areas with high-curvature ridges, such as the areas near singular points. Compared with the two kind methods mentioned above, gradient-based methods are more accurate and subtle, and therefore become one of the most popular methods for the computation of fingerprints’ orientation fields. However, they are sensitive to noise.
For overcoming the defect of gradient-based methods, [4] proposed a hierarchical scheme to dynamically adjust the estimation windows, they introduced a concept of consistence, which means the deviation between the current block orientation and other blocks orientation around it. If the consistency level is above a certain threshold, then the current block orientation is re-estimated at a lower resolution level until it is under a certain level. Wang et al. [5] also proposed a weighted averaging method, the basic idea is to conduct redundant estimation for each target block, following this idea, they design a weighted averaging scheme operated on the target blocks directly. Both of the two methods have made good improvements, but for poor fingerprint images, especially for images with large noise areas, their results are still not very satisfying.
In this paper, we aim to propose a new gradient-based method for the computation of fingerprints’ orientation fields. Comparing with the previously proposed gradient-based approaches, our method will not only possess the advantage of high accuracy, but also be more robust against noise and be capable of predicting.
Section snippets
Related work and analysis
In this section, we will focus on the discussion about the basic gradient-based method introduced by Kass and Witkin [3], which has been adopted by many researchers, such as [14], [15], [16], [17], [18]. The basic gradient-based approach estimates fingerprints’ orientation fields with the following hypothesis: within a limited block area, the orientations of all pixels should almost be the same, so we can employ the block-orientations to replace the pixel-orientations. The reason for using
The combined method for the computation of fingerprints’ orientation fields
As for fingerprints, the changes of orientation patterns are quite smooth and continuous except for the areas around singular points. In this paper, we call ‘smooth areas’ for the areas with smooth orientation patterns, and ‘non-smooth areas’ for the areas around singular points. The prior proposed gradient-based methods calculate fingerprint orientation fields with just single size block, but for a fingerprint image, there consists different orientation patterns. If we take a large block for
Experiments
Two experiments are carried to test the performance of our method. Experiment I aims to directly test the performance of our method in quantitative and qualitative ways; experiment II is designed to evaluate the influence of our method on matching performance.
As the weighted averaging method is more robust against noise than other gradient-based methods [5], in experiment I and experiment II, we both compare the results of our method with the weighted averaging method.
Conclusions
In this paper, we propose a gradient-based combined method for the computation of fingerprints’ orientation field. Compared with the previously proposed gradient-based method, our main contributions include:
(1) Proposing a method which combines the orientation fields calculated by using different size blocks to overcome the defect caused by using just single one.
(2) Proposing an iteration based method to predict the orientations within large noisy areas.
All experiments prove that: compared with
Yuan Mei, is currently working towards the Ph.D. degree in the school of computer science and technology, Nanjing University of Science & Technology. His research interests include fingerprint recognition, image processing, and pattern recognition.
References (20)
- et al.
Analyzing orientation patterns
Comput. Vis. Graph. Image Process.
(1987) - et al.
Fingerprint classification
Pattern Recogn.
(1996) - et al.
A nonlinear orientation model for global description of fingerprints
Pattern Recogn.
(1996) - et al.
A combination model for orientation field of fingerprints
Pattern Recogn.
(2004) - et al.
Constrained nonlinear models of fingerprint orientations with prediction
Pattern Recogn.
(2006) - et al.
Adaptive flow orientation based feature extraction in fingerprint images
Pattern Recogn.
(1995) - et al.
Localization of corresponding points in fingerprints by complex filtering
Pattern Recogn. Lett.
(2003) - et al.
Orientation feature for fingerprint matching
Pattern Recogn.
(2006) - et al.
The history and development of fingerprinting
- Federal Bureau of Investigation, The Science of Fingerprints: Classification and Uses, U.S. Government Printing Office,...
Cited by (24)
A survey of fingerprint classification Part II: Experimental analysis and ensemble proposal
2015, Knowledge-Based SystemsPassword hardened fuzzy vault for fingerprint authentication system
2014, Image and Vision ComputingFingerprint reference point detection for image retrieval based on symmetry and variation
2012, Pattern RecognitionA systematic gradient-based method for the computation of fingerprint's orientation field
2012, Computers and Electrical EngineeringCitation Excerpt :Compared with the previously proposed gradient-based approaches, our method will not only possess the advantage of high accuracy, but also be more robust against noise and be capable of predicting. In this section, we simply introduces the classical gradient-based method which has been adopted by many researchers, such as [5,13–16]. It mainly contains the following three steps:
Frequency domain regularization of d-dimensional structure tensor-based directional fields
2011, Image and Vision Computing
Yuan Mei, is currently working towards the Ph.D. degree in the school of computer science and technology, Nanjing University of Science & Technology. His research interests include fingerprint recognition, image processing, and pattern recognition.
Huaijiang Sun, is the professor and doctor supervisor of Nanjing University of Science and Technology, membership of China Computer Federation. His main interests include image processing, and pattern recognition.
Deshen Xia, is the professor of Nanjing University of Science and Technology and doctor supervisor, His main interests include image processing, and pattern recognition.