指纹是当前应用最为广泛的生物识别特征，然而由于老化、磨损等造成的指纹残缺往往导致识别失败，据统计人群中有3-6%的指纹不能用。真皮指纹与汗腺作为皮下生物特征，不易受表皮状态影响，具有高防伪性能。本项目面向残缺指纹的识别问题，对真皮指纹与汗腺识别进行研究。针对真皮指纹和表皮指纹异源采集的差异问题，提出软硬件结合的指纹配准与基于环状生成对抗网络的图像风格转化，消除采集方式不同带来的分辨率、对比度、脊线粗细等差异的影响，从而建立统一的指纹质量评价体系。针对残缺指纹的修复问题，设计基于质量最优化的指纹融合和基于汗腺连接的脊线重建，弥补残缺指纹的信息。最后，提出基于深度卷积神经网络的指纹与汗腺描述符，鲁棒的描述残缺指纹特征用于识别。本项目的研究为残缺指纹识别提供新的解决思路，有望提高指纹识别的鲁棒性和可靠性，具有重要的研究意义和应用价值。

Fingerprint is the most widely used biometric feature at present. Its incompleteness caused by aging and wear often leads to recognition failure. According to the statistics, 3-6% of the fingerprints in the population cannot be used. As subcutaneous feature, internal fingerprints and sweat glands are not easily affected by epidermal status and have high anti-counterfeiting capability. Aiming at the repair and recognition of incomplete fingerprints, this project studies the internal fingerprints and sweat glands. To remove the inconsistency caused by heterologous acquisition of internal and external fingerprints, a software and hardware based fingerprint registration method, and a circular generative adversarial network based image style conversion method are proposed to adjust the resolution, contrast, ridge thickness and so on. Consequently, a unified quality evaluation system is established for internal and external fingerprints. To improve the quality of the incomplete fingerprints, a repair method is presented including quality optimization based fingerprint fusion, and sweat gland connection based ridge reconstruction. In the last, the fingerprint and sweat gland descriptors are established based on deep convolution neural network for robust recognition of the incomplete fingerprints. This research provides a new idea to solve the problem of incomplete fingerprint identification, and consequently improves the robustness and reliability of fingerprint identification, thus has important theoretical significance and application value.