指纹质量评估及匹配算法研究

发布时间：2018-10-08 21:51

【摘要】：指纹特征具有终生不变和高度区分的天然优势,且指纹识别具有易于实施的工程特性,因此使得指纹识别技术快速普及应用于各类场合。但是指纹识别在实际应用中也存在一些局限性,由于指纹采集设备自身的问题、手指本身的问题或采集不规范等原因,会造成指纹图像质量较差。即便后续的指纹预处理可以进行一定程度的修复,但质量仍然不高,会严重影响匹配识别效果。因此对指纹进行预处理前合适的指纹图像质量评判显得尤为重要。而且常规指纹识别系统多采用单一的匹配策略,识别性能不稳定,匹配效率有限,需要对匹配策略的流程进行深入优化创新。本论文在参考阅读众多文献归纳分析的基础上结合对课题的认真思考,从以下两方面进行了研究和创新:首先本文针对当前指纹识别过程中由于指纹质量问题导致识别性能下降严重的问题,构建了基于数据融合的指纹质量评估方法。本文将通过对多个决策结果进行"取长补短"的融合,有效提高了指纹判断的效果。具体说来,通过对采集到的指纹图像观察其质量问题并进行归纳,第一步得到指纹图像的五个维度的单项评价质量分数,第二步运用有监督学习的分类算法构建多级融合框架对五个质量分数进行多次融合得到一个最终的质量评价分数。通过实验可知,将多个指标信息进行融合之后其判断准确率可显著提升,进而证明了该算法的可效性和实用性。其次针对采用单一的指纹匹配算法识别性能稳定性不高的问题,本文设计了基于级联决策的指纹匹配算法,运用逐次对指纹中心点匹配、指纹中间区域匹配和指纹全局比对三级级联匹配的综合算法,可以有效提高比对算法的识别性能。对于经过配准后的两枚指纹,首先进行指纹中心点比对,若匹配成功则认为是同一人;若失败则继续对指纹中间区域比对,若成功则认为两指纹是同一人;若失败则继续进行指纹的全局比对,若成功则认为是同一人,若失败则不是同一人。此外,为了证明指纹质量评估对于指纹匹配的必要性,本文设计了不同质量的指纹集,通过实验首先验证了指纹质量判断对于指纹匹配的影响很直接,其次验证了与其他匹配方法相比,本文提出的基于级联决策的匹配方法在不同质量的指纹集中依然有着较好的性能。
[Abstract]:Fingerprint features have the natural advantages of lifetime invariance and high discrimination, and fingerprint identification has the engineering characteristics that are easy to implement, so fingerprint identification technology is rapidly applied in various situations. However, fingerprint recognition also has some limitations in practical application. Due to the problems of the fingerprint acquisition equipment itself, the problem of finger itself or the non-standard acquisition, the quality of fingerprint image will be poor. Even though the subsequent fingerprint preprocessing can be repaired to some extent, the quality is still not high, which will seriously affect the matching and recognition effect. Therefore, it is very important to evaluate the fingerprint image quality before the fingerprint preprocessing. Moreover, the conventional fingerprint identification systems usually adopt a single matching strategy, the identification performance is unstable, the matching efficiency is limited, and the process of matching strategy needs to be optimized and innovated deeply. Based on the analysis of many literatures, this paper combines with the serious thinking of the subject. The following two aspects are studied and innovated: firstly, a fingerprint quality assessment method based on data fusion is proposed to deal with the problem that the fingerprint quality problem leads to the serious degradation of fingerprint quality in the current fingerprint identification process. In this paper, the effect of fingerprint judgment is improved by the fusion of multiple decision results. Specifically, by observing and summarizing the quality problems of the collected fingerprint images, the first step is to obtain the single evaluation quality fraction of the five dimensions of the fingerprint image. The second step uses supervised learning classification algorithm to construct multi-level fusion framework to fuse five mass fractions several times to obtain a final quality evaluation score. The experimental results show that the accuracy of the algorithm can be significantly improved after the fusion of multiple index information, which proves the effectiveness and practicability of the algorithm. Secondly, aiming at the problem that the single fingerprint matching algorithm is not stable, this paper designs a fingerprint matching algorithm based on cascade decision. The recognition performance of the matching algorithm can be improved effectively by the three-level cascade matching algorithm of fingerprint middle region matching and fingerprint global matching. For the two fingerprints after registration, the center point of the fingerprint is first compared, if the match is successful, it is considered the same person, if the failure continues to compare the middle region of the fingerprint, if successful, the two fingerprints are considered to be the same person. If the failure continues to carry on the fingerprint global comparison, if successful, thinks is the same person, if the failure then is not the same person. In addition, in order to prove the necessity of fingerprint quality evaluation for fingerprint matching, this paper designs fingerprint sets with different quality. Secondly, compared with other matching methods, the matching method proposed in this paper still has better performance in different quality fingerprint sets.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】