ShoePrint Analysis and Recognition (SPAR)

在侦探科学中，没有哪一个分支像追踪足迹的艺术那样重要而又被忽视。（夏洛克·福尔摩斯：血字研究）这句话在今天几乎和亚瑟·柯南-道尔爵士1887年写下它时一样正确。2006年1月生效的《警察权力法》的修改将允许以与指纹和DNA证据相同的方式处理鞋类证据。也就是说，警察有权获得所有被捕者的鞋印，并可在国家数据库中保存和搜索这些鞋印。目前，从犯罪现场恢复鞋印的情况在英国警察部队中并不多见，但人们认识到，从盗窃场所恢复鞋印的机会比指纹要多。虽然鞋类证据不像指纹或DNA那样独特或永久，但它可以提供非常有价值的情报，在某些情况下可以提供符合法庭标准的证据。从压痕或标记识别鞋模型的当前实践是通过使用鞋模型的数千个图像的手动注释参考集。这些方法是劳动密集型的，并且容易出错，因为鞋类检查员可以采用不同的编码方案。我们拟在这项计划中发展一套全自动化系统，以便迅速和稳健地将鞋模分类，供警方羁押室使用，并发展工具，以协助对现场痕迹进行法证检验，以及提供可予审计的证据标准。我们现有的可行性研究已经证明了开发基于当代图像特征检测器和鲁棒特征描述符的自动印象分类系统的能力。幸运的是，大多数罪犯都穿运动鞋，外底图案的复杂性导致了数千个可识别的特征。通过将一些查询印象的特征与从参考集中的图像导出的特征相关联，我们可以获得可能的候选模型的短列表;然后采用更密集的图形模型来识别特定模型。该项目将以这些概念为基础，以不同的媒体格式制作鞋印和现场标记，改进用于非常大的参考集的技术，科普部分和磨损的标记，提供将嫌疑人的鞋与匹配的现场标记联系起来的工具，并与警方和其他组织合作，进一步将这一技术应用于实际操作。

There is no branch of detective science that is so important and so much neglected as the art of tracing footsteps (Sherlock Holmes: A Study in Scarlet)This quote remains almost as true today as it was when Sir Arthur Conan-Doyle penned it in 1887. Changes in Police Powers, which came into force in January 2006, will permit footwear evidence to be treated in the same manner as fingerprint and DNA evidence. Namely, that the Police have the right to acquire impressions of all arrestees' footwear and that they can be held and searched on a National database. Currently the recovery of footwear marks from crime scenes has been patchy across the individual UK Police Forces but it is recognized that there is more chance to recover footwear marks from burgled premises than fingerprint marks. Though footwear evidence is not as unique or as permanent as fingerprints or DNA, it can provide very valuable intelligence and in some cases evidence to courtroom standards. Current practice in identifying shoe model from impressions or marks is through the use of manually annotated reference sets of many 1000s of images of shoe models. Such methods are labour intensive and prone to error as footwear examiners can employ different coding schemes. We intend in this project to develop a full automatic system for the rapid and robust classification of shoe model for use in Police custody suites and to develop tools to assist in the forensic examination of scene marks and provide auditable evidential standards. Our existing feasibility study has demonstrated the ability to develop automatic impression classification systems based on the use of contemporary image feature detectors and robust feature descriptors. Luckily most criminals wear trainers and the complexity of the outsole patterns results in many 1000s of identifiable features. By correlating the feature of some query impression to those derived from images in the reference set, we can obtain a short-list of likely candidate models; more intensive graphical models are then employed to identify the specific model. This project will build upon these concepts for different media formats of footwear impressions and scene marks, refine techniques for very large reference sets, cope with partial and scuffed marks, provide tools to link a suspect's shoes to the matching scene marks, and to work with the Police and other organizations to further the application of this technology into operational practice.