针对刑侦分析的低质体毛图像毛发模式研究

Forensic analysis based on image is a popular and difficult research area in forensic security and pattern recognition. With the popular of camera device, it is very easy to get data from the digital media. Identifying criminals and victims in these data can be very challenging and useful. The criminals and victims who cover themselves and their faces or tattoos are not observable can be masked gunmen, paedophiles, and victims in child pornographic images. Though skin marks and blood vessel patterns hidden in color images have been proposed to address this problem, they have different limitations. Blood vessel patterns are not suitable for subjects with high concentration of body fat or hair follicle , melanin and skin marks are not suitable for the cases where only low resolution images are available. Medical research results have implied that androgenic hair patterns are a stable biometric trait and have potential to overcome the weaknesses of skin mark patterns and blood vessel patterns. This project aims at developing a new method based on androgenic hair patterns for forensic analysis. To the best knowledge of the authors, no one studied androgenic hair patterns for criminal and victim identification before..Because the varations of weather and light , uncontrol of the capture distance and viewpoints et al,normally the evidence image is low resolution,low quality , distore by variations of viewpoints and poses. Considering low quality and variations are unavoidable in real cases, the project will study on the representation of low quality androgenic hair patterns, alignment problem and principal curves classification to model the framework and algorithm of androgenic hair patterns filter, mutli-feature fusion and alignment for implementing the analysis and matching based on androgenic hair patterns for criminal and victim identification. The androgenic hair patterns in low resolution will be a novel method for forensic analysis.

基于图像的刑侦分析是刑侦、模式识别领域的研究热点和难点。摄像设备的普及，为刑侦分析提供了大量有利的图像、视频数据，通过这些数据来认定犯罪者是非常具有意义和挑战性的。特别是在暴力游行、恐怖袭击、儿童性侵等案件中，犯罪者刻意遮盖，其脸部、纹身等常常不可见。虽然皮肤标志和静脉血管图的认定方法已经提出来解决这个问题，但在普通图像中直接获取它们是非常困难的，且痣、皮肤标记及毛囊在非高清图像中无法工作。针对以上问题，项目提出采用体毛毛发模式这种新的生物特征技术进行刑侦分析。.针对实际环境中证物图像易受拍摄光线、天气、拍摄距离及视角等不利因素影响而存在着低分辨率、低质量、视角变化、姿势变化等问题，本项目通过对低质毛发模式表征、视角无关及姿势无关的特征配准、主曲线分类的研究，构造基于毛发滤波器及多特征融合、配准的毛发模式匹配框架和新算法，实现低质图像毛发模式分析及匹配，力求为司法认定提供更加可靠的方法。

基于图像的刑侦分析是刑侦、模式识别领域的研究热点和难点。摄像设备的普及，为刑侦分析提供了大量有利的图像、视频数据，通过这些数据来认定犯罪者是非常具有意义和挑战性的。特别是在暴力游行、恐怖袭击、儿童性侵等案件中，犯罪者刻意遮盖，其脸部、纹身等常常不可见。虽然皮肤标志和静脉血管图的认定方法已经提出来解决这个问题，但在普通图像中直接获取它们是非常困难的，且痣、皮肤标记及毛囊在非高清图像中无法工作。针对以上问题，项目提出采用体毛毛发模式这种新的生物特征技术进行刑侦分析。. 针对实际环境中证物图像易受拍摄光线、天气、拍摄距离及视角等不利因素影响而存在着低分辨率、低质量、视角变化、姿势变化等问题，项目围绕毛发模式表征、分析、匹配三方面进行了研究。研究了基于毛发模式滤波器的多特征识别方法；直接表征了毛发的密度特征，提出了毛发密度分类器，将毛发密度分为高、中、低三个等级，并基于毛发密度分类器修正了动态格子系统存在的误配准；针对视角和姿态变化问题，借鉴了地理等高线的概念，提出了基于皮肤纹理趋势分布的特征表征方法，并设计了滑块系统以增强匹配的准确度、提高算法的鲁棒性。对于视角偏转较大、姿势变化较大的情况，提出了新的腿部视角变换模型。项目在较低的代价下解决了低质图像因视角变化、姿势变化造成的误配准问题，实现了低质图像毛发模式分析及匹配。此外，基于皮肤纹理趋势分布及新的匹配策略研究了遮挡问题；为提高识别速度，采用卷积神经网络实现了基于腿部图像及其毛发模式区分男女性别的方法。