A Study Of Fingerprint Absolute Pre-alignment Approach

As a valuable research topic in the field of fingerprint recognition,the fingerprint absolute pre-alignment technique is a vital means of accelerating the matching procedure for systems with large database.Recently,the increasing demand of the automated fingerprint identification systems activates the research on how to improve the systems' responding speed to the real-time level.One of the advisable solutions is to utilize more advanced hardware facilities.However,the upgrade and maintenance of these facilities can be extreme expensive.An alternative economical way is to speed up the matching algorithms where the fingerprint absolute pre-alignment method can be used.To reach this goal,the fingerprint absolute pre-alignment technique tries to build a unified coordinate system for each fingerprint image so as to eliminate the effects of transformation and rotation.As a result,the matching time for each pair fingerprint images is gracefully reduced.Unfortunately,the process of building the unified coordinate system itself can be time-consuming.Moreover,the stability and accuracy of the built coordinate system has not yet satisfied the demands of most applications.To solve the above problems,this work proposed some solutions in the aspects of building the original point and direction of the unified coordinate system.The work can be concretely divided into the following three parts:(1)We propose a walking-based fingerprint singular points detection algorithm Fingerprint singular point,an importance kind of reference points,can serve as the original point of the unified coordinate system in most fingerprints.Based on the analysis of the local orientation field around singular point simulated using the zero-pole model and the orientation field estimated from real fingerprint images,the proposed walking algorithm deduces a new directional field termed Walking Directional Field(WDF).From any point on WDF,we can directly walk to the corresponding singular point by following the walking direction.The algorithm succeeds to avoid scanning the fingerprint image,so it is dramatically efficient even for pixel-wise orientation field.Furthermore,the walking algorithm can also remove spurious singular points by repeatedly walking in the neighborhood of the candidate singular point.By adding an angle correction term into the calculation formula of WDF,the walking algorithm can deal with fingerprint images with different rotation angles.The experimental results in a public fingerprint dataset validate the efficiency of the walking algorithm.(2)We propose a fast and accurate method for detecting fingerprint reference point Since no singular points are defined for arch type fingerprints,we propose a method for detecting reference points in all types of fingerprints.Firstly,we make two improvements for the mean-shift based algorithm: 1)use orientation field with a block size of 8×8 instead of 16×16;2)assign a weight to each cross point and use the weighted average point of cross points as the reference point.The enhanced algorithm is much more accurate but also more time-consuming.To improve the efficiency,we revise the walking algorithm to a simplified version which is more efficient and suitable for working on block-wise orientation field.Then the simplified walking algorithm is used to detect the coarse position of the reference point and in the local area the enhanced mean-shift based algorithm is applied to refine the position.By combining these two algorithms,both efficiency and accuracy are improved compared with the single algorithms.Experimental results on two popular publicly available datasets show the proposed method is faster and more accurate than two state-of-the-art methods.(3)We propose a reference point detection algorithm based on parallelism and symmetry After detecting stable reference point,the reference direction must be estimated so as to build the complete unified coordinate system.The proposed method combines the global stability of the parallelism-based method and the local accuracy of the symmetry-based method.It achieves higher accuracy without supernumerary time cost.Compared with the parallelism-based method,our method is more stable for those fingerprints where detected reference points shift away from the true position of corresponding core points.We tested the algorithm on the public dataset FVC2000 DB2 a and the results proved the accuracy of the proposed method.