基于视觉感知模型的指纹图像质量评价算法

Fingerprint image quality evaluation algorithm based on visual perception model

指纹在刑事侦查和法庭科学领域被称为物证之首,在打击犯罪、维护社会稳定中发挥着十分重要的作用。指纹鉴定的结论是有效证据形成和刑事案件侦破的关键,而指纹图像质量直接影响指纹鉴定结论的准确性和可靠性,因此精准客观的指纹图像质量评价算法是辅助指纹专家进行指纹鉴定必不可少的工具。目前,美国国家标准与技术研究院研发的NFIQ2.0指纹图像质量评价算法得到国内外专家学者的广泛关注,但该算法质量评价结果与指纹专家的质量评价结果存在较大偏差,且缺乏对指纹图像局部区域的质量评价,所以无法满足刑事侦查和法庭科学领域指纹鉴定任务的需要。基于此,将指纹图像质量评价问题拓展到质量空间上,学习指纹专家对纹线局部区域的质量感知策略,提出基于专家感知的质量评价算法。实验结果表明本文算法对指纹图像整体质量评价的结果能够与指纹专家的质量评价结果保持一致,并且符合刑事侦查和法庭科学的应用场景。此外,进一步在国际公开指纹数据集上与NFIQ2.0算法进行对比实验,结果表明本文算法的质量评价分数更为合理,能够有效降低指纹匹配算法的拒识率。

Fingerprint is known as the first material evidence in the field of criminal investigation and forensic science,and plays a very important role in combating crime and maintaining social stability.The conclusion of fingerprint identification is the key to the formation of effective evidence and the detection of criminal cases,and the quality of fingerprint image directly affects the accuracy and reliability of fingerprint identification conclusions,so accurate and objective fingerprint image quality evaluation algorithm is an indispensable tool to assist fingerprint experts in fingerprint identification.At present,the NFIQ2.0 fingerprint image quality evaluation algorithm developed by the National Institute of Standards and Technology(NIST)has attracted extensive attention from experts and scholars at home and abroad.However,the quality evaluation results of this algorithm deviate greatly from the quality evaluation results of fingerprint experts,and lack of quality evaluation for local areas of fingerprint images.Therefore,it can not meet the needs of criminal investigation and forensic science fingerprint identification tasks.In this paper,the problem of fingerprint image quality evaluation is extended to the quality space,and the quality evaluation algorithm based on expert perception is proposed by learning the quality perception strategy of fingerprint experts for the local area of ridges.The experimental results show that the overall quality evaluation results of the proposed algorithm are consistent with the quality evaluation results of fingerprint experts,and accord with the application scenarios of criminal investigation and forensic science.In addition,this paper further compared with the NFIQ2.0 algorithm on several international public fingerprint data sets,and the results show that the quality evaluation score of the proposed algorithm is more reasonable,and can effectively reduce false non-match rate(FNMR)of the fingerprint matching algorithm.