现场指纹自动处理与识别若干关键问题研究

Latent fingerprints refer to the prints lifted from surfaces of objects in criminal scene. Compared to rolled and plain fingerprints (or exemplar fingerprints), which are acquired in an attended mode, latents are typically of poor quality in terms of ridge structure and complex background noise, and are usually partial (incomplete) and with large non-linear distortion. Current latent fingerprint recognition systems rely on manually marked features and manual intervention is also needed in the verification. In this proposal, we plan to work on automatic processing and matching of latent fingerprints. First, dictionary learning and sparse coding based approach will be proposed for latent segmentation, enhancement and feature extraction. A number of high quality fingerprints are used for fingerprint patch dictionary learning, which will be used to reconstruct latent fingerprint patches. The fingerprint quality is then defined as the similarity between latent patches and their reconstructions. The orientation and frequency fields, which are used for enhancement, are then extracted from the reconstructed patches. Dictionary based method will be also used to reconstruct the whole orientation field from the incomplete latent fingerprints, which can be used to reduce false reject rate. Local features around minutiae will be learnt and fused to increase minutiae distinctiveness. Local matching and propagation and tensor based method will be proposed to ensure the topological structure of matched minutiae.

现场指纹是从犯罪现场提取的指纹图像。与主观配合下采集的滚动指纹和平板指纹相比，现场指纹不仅脊线纹理质量差，背景复杂，背景噪声甚至与指纹纹理重叠，而且提取出来的指纹通常不完整，具有很大形变。目前的现场指纹识别系统都依赖手动标定的特征，在认证过程中也需要人工的参与。在本项目拟研究现场指纹的自动处理和识别。我们拟采用字典学习和稀疏表示的方法对现场指纹图像进行分割，增强和特征提取，大量高质量的指纹图像块用来学习得到字典，对现场指纹块采用字典和稀疏表示的方法对其重构。指纹图像的质量就可以定义为现场指纹图像块与其重建块之间的相似度，而用来增强的指纹方向场和频率场则从其重建块中提取。字典学习的方法还将用来从不完整的现场指纹方向场中重建完整的方向场以降低错误错误接受率。本项目还将研究细节点局部特征的提取和融合以提高细节点的区分性，以及局部细节点匹配和基于张量的图匹配来保证匹配细节点的拓扑结构。

现场指纹是从犯罪现场提取的指纹图像。为了解决现场指纹脊线纹理质量差，背景复杂，背景噪声甚至与指纹纹理重叠的问题，提升现场指纹的自动处理和识别性能，课题组针对目前现场指纹识别的重难点，进行了大量的研究。在现场指纹细节点提取方面，提出了基于指纹一级特征的算法；在指纹增强方面，基于卷积神经网络的算法完成现场指纹方向场的估计，并基于滤波器组进行现场指纹增强，以保证可靠有效地增强现场指纹图像并提取指纹特征；在指纹匹配领域，除了提出了基于包络的方法和基于纹理模板的指纹识别算法。除了现场指纹识别领域，还在特殊指纹识别、指纹3-D建模与合成及指纹检索领域做出了一定成果。课题组在国内外会议和期刊上发表了学术论文6篇，授权专利1项，技术成果转移1项，受理专利10项，获得陕西省科学技术奖二等奖1项，出版指纹相关教材1部。我们的研究成果为现场指纹识别技术的进步做出了一定的贡献，促进了指纹识别技术的发展。