基于指纹的婴幼儿身份快速识别

Infant and newborn recognition has critical applications in national household registration management, prevention of child abduction and hospital newborn management. In current practice, although the DNA based identification technology can accurately recognize infants and newborns, however, it can not satisfy the need to fast and real-time requirements in real world scenario, and the price is very expensive. With the development of software and hardware technology, biometric recognition technique has gained remarkable improvement. Fingerprint is a well-known biometric modality with reliable and high discriminative features. In this project, we take fingerprint as the main biometric recognition modality, explore the internal invariant characteristics during growth of infant and newborn fingerprint, and based on these invariant characteristics, research on infant and newborn recognition system which robust to growth. We will focus on the problems of fingerprint image quality evaluation, fingerprint growth model, accurate matching algorithm and online template update, provide a fast, accurate and non-expensive identification solution to the special fast growing group of infant and newborn.The results of this project will provide a solid technology foundation to improve national household registration, prevent child abduction and improve hospital newborn management, etc.

婴幼儿身份认证在国家户籍管理、打击拐卖儿童犯罪、医院新生儿管理等领域具有重要的应用。目前，基于DNA验证的身份认证技术虽然能够对婴幼儿进行准确识别，但是无法满足快速、实时的实际应用需求，且价格十分昂贵。随着软硬件技术水平的提高，生物特征识别技术近年来得到了迅猛的发展。指纹是已知公认的稳定且区分性高的生物特征模态，本项目以指纹做为婴幼儿生物特征识别的主要模态，探索婴幼儿手指表皮脊线纹理的成长内在不变性特点，并以这些内在不变性特点为识别的依据，研究对成长鲁棒的婴幼儿指纹识别系统，着重解决婴幼儿指纹质量评估、指纹的成长形变模型、指纹的精确匹配和模板在线更新等关键技术难点问题，为婴幼儿这一处在特殊发育时期的群体提供一套快速、准确、廉价的身份识别解决方案，为完善国家户籍管理制度、打击拐卖儿童犯罪、医院新生儿身份管理等重要应用提供坚实的技术保障。

婴幼儿身份认证在国家户籍管理、打击拐卖儿童犯罪、医院新生儿管理等领域 具有重要的应用。目前，基于 DNA 验证的身份认证技术虽然能够对婴幼儿进行准确识别，但 是无法满足快速、实时的实际应用需求，且价格十分昂贵。婴幼儿处在一个快速发育的人生阶段，手指生长速度很快，皮肤的拉升和扭曲严重，同时，婴幼儿的指纹纹理很密，采集难度较高。本项目研究了基于指纹的婴幼儿身份识别问题，取得丰富的研究成果。首先，在项目的资助下，我们采集了高分辨婴幼儿生物特征图像数据库，包含了100多例婴幼儿的指纹、掌纹、脚印数据，我们跟踪采集了婴幼儿从满月到9个月大的数据生物特征；其次，我们提出了基于siamese网络的婴幼儿指纹和脚印识别算法，并采用迁移学习技术解决数据量不够的问题，显著提高了婴幼儿的身份识别准确率；第三，我们研究了婴幼儿指纹细节点与方向场之间的关联性，从理论上证明了由方向场恢复指纹细节点的可能性，并从实验验证了指纹重构的精度；第四，我们提出了基于深度学习的婴幼儿指纹方向场提取算法，该算法在FVC-OnGoing指纹识别竞赛网站中夺冠。