Research On Sensor Interoperability Problem Of The Fingerprint Image Segmentation

Because people's physiological features have the characteristics of easy to carry, not lost, not easy to imitate, high complexity and so on, biometric identification technology has become one of the primary means of identity authentication in information society. Fingerprint recognition technology is the earliest and most widely used technology among all the biometrics. Automated fingerprint identification system is an intelligent automated system that uses computer-based high speed computing power, a variety of pattern recognition technology, digital image processing, computer vision technology, morphological techniques, wavelet analysis and machine learning and other technologies.In the automated fingerprint identification system, fingerprint image segmentation is a very important part. In recent years, sensor interoperability has got a lot of attention in the fingerprint identification field. Now, most automated fingerprint identification systems are designed for one particular fingerprint sensor. When system deals with fingerprint images from different types of sensors, performance is often in different degrees of decline in steps of segmentation, enhancement, matching and so on.This is because fingerprint images collected by different sensors often have different image qualities, resolutions and gray levels.A good fingerprint image segmentation method that owns sensor interoperability should be able to adapt to fingerprint images collected by different sensors and be reliable to achieve segmentation on the premise of not modifying and not doing configuration parameters.First of all, this thesis proposes the existence of sensor interoperability in fingerprint image segmentation.In order to better illustrate the problem's influence to fingerprint image segmentation, we design two groups of fingerprint image segmentation experiments.Through analysis, we reveal the existence of sensor interoperability in most current fingerprint image segmentation methods and propose the need for researching segmentation methods that own sensor interoperability. Then, we propose two fingerprint image segmentation methods that have characteristic of sensor interoperability. The first method is based on K-means algorithm. The three block-based features of direction consistency, average gray and gray variance are extracted. Blocks are clustered by using K-means clustering algorithm. It can distinguish foreground blocks and background blocks from one image. By using morphological post-processing, fingerprint image segmentation is implemented ultimately. The second method is based on LNP algorithm.First, the three block-based features of direction consistency, aggregation and gray pixel variance are extracted. Then,the blocks are divided into foreground area and background area by using LNP algorithm.Finally, by using morphological post-processing, the fingerprint image segmentation is implemented.The experimental results in a number of databases show that not only the two methods have good performance and can implement fingerprint image segmentation accurately, but also they own sensor interoperability. The two methods are more suitable for the Internet applications environment that uses multi-type fingerprint sensors. This thesis focuses on researching sensor interoperability in the fingerprint image segmentation.The organic combination of research results and existing methods to solve sensor interoperability has positive scientific significance to promote the further solving sensor interoperability. It also promotes the further network application of fingerprint identification technology and it has an important reference value to the solution of similar problems in other biometric identification algorithms.