Automatic forensic face recognition from digital images.

Digital image evidence is now widely available from criminal investigations and surveillance operations, often captured by security and surveillance CCTV. This has resulted in a growing demand from law enforcement agencies for automatic person-recognition based on image data. In forensic science, a fundamental requirement for such automatic face recognition is to evaluate the weight that can justifiably be attached to this recognition evidence in a scientific framework. This paper describes a pilot study carried out by the Forensic Science Service (UK) which explores the use of digital facial images in forensic investigation. For the purpose of the experiment a specific software package was chosen (Image Metrics Optasia). The paper does not describe the techniques used by the software to reach its decision of probabilistic matches to facial images, but accepts the output of the software as though it were a 'black box'. In this way, the paper lays a foundation for how face recognition systems can be compared in a forensic framework. The aim of the paper is to explore how reliably and under what conditions digital facial images can be presented in evidence.     

The mask detection technology for occluded face analysis in the surveillance system

The surveillance systems have been widely used in automatic teller machines (ATMs), banks, convenient stores, etc. For example, when a customer uses the ATM, the surveillance systems will record his/her face information. The information will help us understand and trace who withdrew money. However, when criminals use the ATM to withdraw illegal money, they usually block their faces with something (in Taiwan, criminals usually use safety helmets or masks to block their faces). That will degrade the purpose of the surveillance system. In previous work, we already proposed a technology for safety helmet detection. In this paper, we propose a mask detection technology based upon automatic face recognition methods. We use the Gabor filters to generate facial features and utilize geometric analysis algorithms for mask detection. The technology can give an early warning to save-guards when any "customer" or "intruder" blocks his/her face information with a mask. Besides, the technology can assist face detection in the automatic face recognition system. Experimental results show the performance and reliability of the proposed technology.     

In‐Depth Analysis of Computer Memory Acquisition Software for Forensic Purposes

The comparison studies on random access memory (RAM) acquisition tools are either limited in metrics or the selected tools were designed to be executed in older operating systems. Therefore, this study evaluates widely used seven shareware or freeware/open source RAM acquisition forensic tools that are compatible to work with the latest 64‐bit Windows operating systems. These tools' user interface capabilities, platform limitations, reporting capabilities, total execution time, shared and proprietary DLLs, modified registry keys, and invoked files during processing were compared. We observed that Windows Memory Reader and Belkasoft's Live Ram Capturer leaves the least fingerprints in memory when loaded. On the other hand, ProDiscover and FTK Imager perform poor in memory usage, processing time, DLL usage, and not‐wanted artifacts introduced to the system. While Belkasoft's Live Ram Capturer is the fastest to obtain an image of the memory, Pro Discover takes the longest time to do the same job.     

法庭语音学

本文对法庭语音学进行了研究综述,主要介绍了该学科的核心内容：说话人鉴定。在实际办案中,当未找到嫌疑人,只有犯罪分子的检材语音时,可以使用说话人画像/说话人分类技术。若没有犯罪分子的录音证据时,可以让受害人和证人进行说话人的听觉辨认。具体的辨认形式有两种：对熟人辨认和对陌生人辨认,在对陌生人辨认时可以采用语音辨认的方法进行。当检材语音和样本语音都齐备的时候,法庭语音分析专家就可以对二者进行比对检验了。目前语音比对分析涉及到的问题和领域有：基于贝叶斯方法的法庭推理和似然比计算、共振峰频率的测量应用、非解析感知与样例理论、法庭说话人自动识别以及不同方法的综合应用等。     

Portable high‐resolution automated 3D imaging for footwear and tire impression capture

The forensic science community raised the need for improved evidence recognition, collection, and visualization analytical instrumentation for field and laboratory use. While the 3D optical techniques for imaging static objects have been extensively studied, there is still a major gap between current knowledge and collecting high‐quality footwear and tire impression evidence. Among optical means for 3D imaging, digital fringe projection (DFP) techniques reconstruct 3D shape from phase information, achieving camera‐pixel spatial resolution. This paper presents a high‐resolution 3D imaging technology using DFP techniques dedicated to footwear and tire impression capture. We developed fully automated software algorithms and a graphical user interface (GUI) that allow anyone without training to operate for high‐quality 3D data capture. We performed accuracy evaluations and comparisons comparing with the commercial high‐end 3D scanner and carried out qualitative tests for various impressions comparing with the current practices. Overall, our technology achieves similar levels of accuracy and resolution with a high‐end commercially available 3D scanner, while having the merits of being (1) more affordable; (2) much easier to operate; and (3) more robust. Compared with the current practice of casting, our technology demonstrates its superiority because it (1) is non‐destructive; (2) collects more evidence detail than casts, especially when an impression is fragile; (3) requires less time and money to collect each piece of evidence; and (4) results in a digital file that can easily be shared with other examiners.     

Image conditions for machine-based face recognition of juvenile faces

Machine-based facial recognition could help law enforcement and other organisations to match juvenile faces more efficiently. It is especially important when dealing with indecent images of children to minimise the workload, and deal with moral and stamina challenges related to human recognition. With growth related changes, juvenile face recognition is challenging. The challenge not only relates to the growth of the child’s face, but also to face recognition in the wild with unconstrained images. The aim of the study was to evaluate how different conditions (i.e. black and white, cropped, blur and resolution reduction) can affect machine-based facial recognition of juvenile age progression. The study used three off-the-shelf facial recognition algorithms (Microsoft Face API, Amazon Rekognition, and Face++) and compared the original images and the age progression images under the four image conditions against an older image of the child. The results showed a decrease in facial similarity with an increased age gap, in comparison to Microsoft; Amazon and Face++ showed higher confidence scores and are more resilient to a change in image condition. The image condition ‘black and white’ and ‘cropped’ had a negative effect across all three APIs. The relationship between age progression images and the younger original image was explored. The results suggest manual age progression images are no more useful than the original image for facial identification of missing children, and Amazon and Face++ performed better with the original image.     

Image matching algorithms for breech face marks and firing pins in a database of spent cartridge cases of firearms

On the market several systems exist for collecting spent ammunition data for forensic investigation. These databases store images of cartridge cases and the marks on them. Image matching is used to create hit lists that show which marks on a cartridge case are most similar to another cartridge case. The research in this paper is focused on the different methods of feature selection and pattern recognition that can be used for optimizing the results of image matching.The images are acquired by side light images for the breech face marks and by ring light for the firing pin impression. For these images a standard way of digitizing the images used. For the side light images and ring light images this means that the user has to position the cartridge case in the same position according to a protocol. The positioning is important for the sidelight, since the image that is obtained of a striation mark depends heavily on the angle of incidence of the light. In practice, it appears that the user positions the cartridge case with +/-10 degrees accuracy.We tested our algorithms using 49 cartridge cases of 19 different firearms, where the examiner determined that they were shot with the same firearm. For testing, these images were mixed with a database consisting of approximately 4900 images that were available from the Drugfire database of different calibers.In cases where the registration and the light conditions among those matching pairs was good, a simple computation of the standard deviation of the subtracted gray levels, delivered the best-matched images. For images that were rotated and shifted, we have implemented a "brute force" way of registration. The images are translated and rotated until the minimum of the standard deviation of the difference is found. This method did not result in all relevant matches in the top position. This is caused by the effect that shadows and highlights are compared in intensity. Since the angle of incidence of the light will give a different intensity profile, this method is not optimal.For this reason a preprocessing of the images was required. It appeared that the third scale of the "à trous" wavelet transform gives the best results in combination with brute force. Matching the contents of the images is less sensitive to the variation of the lighting.The problem with the brute force method is however that the time for calculation for 49 cartridge cases to compare between them, takes over 1 month of computing time on a Pentium II-computer with 333MHz. For this reason a faster approach is implemented: correlation in log polar coordinates. This gave similar results as the brute force calculation, however it was computed in 24h for a complete database with 4900 images.A fast pre-selection method based on signatures is carried out that is based on the Kanade Lucas Tomasi (KLT) equation. The positions of the points computed with this method are compared. In this way, 11 of the 49 images were in the top position in combination with the third scale of the à trous equation. It depends however on the light conditions and the prominence of the marks if correct matches are found in the top ranked position. All images were retrieved in the top 5% of the database. This method takes only a few minutes for the complete database if, and can be optimized for comparison in seconds if the location of points are stored in files.For further improvement, it is useful to have the refinement in which the user selects the areas that are relevant on the cartridge case for their marks. This is necessary if this cartridge case is damaged and other marks that are not from the firearm appear on it.     

深度学习在图像识别及骨龄评估中的优势及应用前景

深度学习以及神经网络模型是近年来机器学习及人工智能领域新的研究方向及热点问题。深度学习在图像识别、语音识别应用中已取得了突破性进展,在人脸识别、信息检索等领域也展示出独特优势,得到了广泛应用。骨骼X线图像显示黑白灰不同阶度的变化,具有黑白对比、层次差异的图像特征,基于深度学习在图像识别中的优势,我们将其与骨龄评估研究有机结合,旨在为构建法医学骨龄自动化评估系统提供基础性数据。本文综述了深度学习的基本概念及其网络结构,阐述了近年来深度学习在国内外不同研究领域图像识别中的研究进展,以及深度学习在骨龄评估中的优势及应用前景。     

Whole-face procedures for recovering facial images from memory

Research has indicated that traditional methods for accessing facial memories usually yield unidentifiable images. Recent research, however, has made important improvements in this area to the witness interview, method used for constructing the face and recognition of finished composites. Here, we investigated whether three of these improvements would produce even-more recognisable images when used in conjunction with each other. The techniques are holistic in nature: they involve processes which operate on an entire face. Forty participants first inspected an unfamiliar target face. Nominally 24 h later, they were interviewed using a standard type of cognitive interview (CI) to recall the appearance of the target, or an enhanced ‘holistic’ interview where the CI was followed by procedures for focussing on the target's character. Participants then constructed a composite using EvoFIT, a recognition-type system that requires repeatedly selecting items from face arrays, with ‘breeding’, to ‘evolve’ a composite. They either saw faces in these arrays with blurred external features, or an enhanced method where these faces were presented with masked external features. Then, further participants attempted to name the composites, first by looking at the face front-on, the normal method, and then for a second time by looking at the face side-on, which research demonstrates facilitates recognition. All techniques improved correct naming on their own, but together promoted highly-recognisable composites with mean naming at 74% correct. The implication is that these techniques, if used together by practitioners, should substantially increase the detection of suspects using this forensic method of person identification.     

Configural and featural information in facial-composite images

Eyewitnesses are often invited to construct a facial composite, an image created of the person they saw commit a crime that is used by law enforcement to locate criminal suspects. In the current paper, the effectiveness of composite images was investigated from traditional feature systems (E-FIT and PRO-fit), where participants (face constructors) selected individual features to build the face, and a more recent holistic system (EvoFIT), where they ‘evolved’ a composite by repeatedly selecting from arrays of complete faces. Further participants attempted to name these composites when seen as an unaltered image, or when blurred, rotated, linearly stretched or converted to a photographic negative. All of the manipulations tested reduced correct naming of the composites overall except (i) for a low level of blur, for which naming improved for holistic composites but reduced for feature composites, and (ii) for 100% linear stretch, for which a substantial naming advantage was observed. Results also indicated that both featural (facial elements) and configural (feature spacing) information were useful for recognition in both types of composite system, but highly-detailed information was more accurate in the feature-based than in the holistic method. The naming advantage of linear stretch was replicated using a forensically more-practical procedure with observers viewing an unaltered composite sideways. The work is valuable to police practitioners and designers of facial-composite systems.     

图像降噪和增强对人脸识别系统识别性能的影响

目的研究图像降噪和图像增强方法对人脸识别系统识别性能的影响,以期为人脸识别系统应用过程中的图像处理方法选取提供理论指导和技术方案。方法收集33起人像鉴定领域实际案例中的人脸图像素材,研究以高斯滤波和小波变换为代表的图像降噪技术以及具有边缘保持和小波变换特性的单帧图像超分辨率增强技术对人脸识别系统识别性能的影响,并对不同图像处理方法对人脸识别性能的影响进行量化比较分析。结果本文研究的图像降噪技术均显著提高了人脸识别系统的识别准确性,而图像增强技术虽然提高了人脸图像显示效果,但对人脸识别系统的识别性能无正向促进作用。此外,高斯模糊图像处理的图像降噪方法虽然简单,但与本文研究的其他方法比较,其在人脸识别系统识别性能改善方面效果最显著。结论人脸图像质量对人脸识别系统的识别性能具有显著影响,可以通过图像处理技术改善人脸图像质量进而提高人脸识别系统的识别准确性。其中,图像降噪处理可以显著提高人脸识别系统的识别性能,且比图像增强技术更适合于实际人像鉴定应用中的人脸识别系统识别性能增强。     

Eye‐Catching Eyebrows: Training and Evaluating Humans for Periocular Image Verification

High‐quality periocular images have sufficient variability between people that they can be used for automated recognition. However, there are no standards for training people to compare periocular images in a forensic context. Potential methods for training people to compare eyebrow images are investigated in this article. Pairs of images were presented to volunteers who decided whether the images showed the same person or different people. A control group with minimal instruction and no training responded with 87.6% accuracy. A second group viewed 100 training pairs before the test and achieved 94.2% accuracy. A third group rated specific features in the images and had 0.1% improvement over the second group. Experimental results show that actively comparing at least thirty training images and receiving immediate feedback improves trainees’ performance six to seven percent. If a quantitative quality score is needed, a linear combination of similarity ratings on different features may be used.     

证人的目击面孔心理重建及认知检测

提取目击证人对犯罪嫌疑人的面孔记忆是一项困难的工作,由于证人在识记面孔时有多种影响因素,在提取和再认记忆时又会有各种认知因素的影响,导致面孔心理重建的失真。因此,从认知心理学和实验心理学的角度入手,分析影响心理重建的因素,并用最佳的手段提取记忆,可以使面孔重建的可靠性最大化。认知相关电位在记忆检测领域的日臻完善,使目击证人的再认正确与否有了可靠的参照,从记忆反应的脑电图形分析,可以有效判断证人对列队辨认中的面孔是否存有记忆。     

The ideal subject distance for passport pictures

In an age of global combat against terrorism, the recognition and identification of people on document images is of increasing significance. Experiments and calculations have shown that the camera-to-subject distance - not the focal length of the lens - can have a significant effect on facial proportions. Modern passport pictures should be able to function as a reference image for automatic and manual picture comparisons. This requires a defined subject distance. It is completely unclear which subject distance, in the taking of passport photographs, is ideal for the recognition of the actual person. We show here that the camera-to-subject distance that is perceived as ideal is dependent on the face being photographed, even if the distance of 2m was most frequently preferred. So far the problem of the ideal camera-to-subject distance for faces has only been approached through technical calculations. We have, for the first time, answered this question experimentally with a double-blind experiment. Even if there is apparently no ideal camera-to-subject distance valid for every face, 2m can be proposed as ideal for the taking of passport pictures. The first step would actually be the determination of a camera-to-subject distance for the taking of passport pictures within the standards. From an anthropological point of view it would be interesting to find out which facial features allow the preference of a shorter camera-to-subject distance and which allow the preference of a longer camera-to-subject distance.     

“省管县”:历史与现实之间的观照——中国地方行政层级的优化改革

中国地方行政层级变迁的历史周期表明:不论"二级制",还是"三级制",总会出现某一层级直接管理的下级单位数目过多,此时总会面临上级能否对下级有效管理的难题。我国地方行政层级的改革进程经历了从"市管县"到"省管县"的变迁,而"省管县"的改革模式和改革目标都存在多样的可能。今后的改革要充分考虑到自然环境、管理半径、监督制度和职能转变等诸多因素,提高可操作性。     

基于语言学音系例字的口音自动识别探究

目的以方言语音学为视角,通过计算机分析和选择模型训练数据,探究提高口音自动识别准确率及优化模型训练数据的方法。方法采用音系分析、语音信号处理、数学模型实验和统计的方法,对包含约81400段电子语音的37个方言点(每个方言点约2200段),逐一进行音系提炼、挑选音系例字、电子语音预处理和提取梅尔倒谱系数(MFCC),构建高斯混合模型(GMM)进行口音识别分析。结果从方言语音中提取的音系例字作为训练集(约260字)的识别模型可以较好地完成口音识别任务,对比任选300个例字作为训练集的识别模型,无论是识别准确度还是待测语音数量的要求都有显著优势。结论基于语言学音系例字的口音自动识别方法,已经完成37个方言点高斯混合模型的建立,可以用于口音识别中辅助辨别分析。     

Automated detection of regions of interest in cartridge case images using deep learning

This paper explores a deep-learning approach to evaluate the position of circular delimiters in cartridge case images. These delimiters define two regions of interest (ROI), corresponding to the breech face and the firing pin impressions, and are placed manually or by an image-processing algorithm. This positioning bears a significant impact on the performance of the image-matching algorithms for firearm identification, and an automated evaluation method would be beneficial to any computerized system. Our contribution consists in optimizing and training U-Net segmentation models from digital images of cartridge cases, intending to locate ROIs automatically. For the experiments, we used high-resolution 2D images from 1195 samples of cartridge cases fired by different 9MM firearms. Our results show that the segmentation models, trained on augmented data sets, exhibit a performance of 95.6% IoU (Intersection over Union) and 99.3% DC (Dice Coefficient) with a loss of 0.014 for the breech face images; and a performance of 95.9% IoU and 99.5% DC with a loss of 0.011 for the firing pin images. We observed that the natural shapes of predicted circles reduce the performance of segmentation models compared with perfect circles on ground truth masks suggesting that our method provide a more accurate segmentation of the real ROI shape. In practice, we believe that these results could be useful for firearms identification. In future work, the predictions may be used to evaluate the quality of delimiters on specimens in a database, or they could determine the region of interest on a cartridge case image.     

Computer-based production of comparison overlays from 3D-scanned dental casts for bite mark analysis

Bite mark analysis assumes the uniqueness of the dentition can be accurately recorded on skin or an object. However, biting is a dynamic procedure involving three moving systems, the maxilla, the mandible, and the victim's reaction. Moreover, bite marks can be distorted by the anatomic location of the injury or the elasticity of the skin tissue. Therefore, the same dentition can produce bite marks that exhibit variations in appearance. The complexity of this source of evidence emphasizes the need for new 3D imaging technologies in bite mark analysis. This article presents a new software package, DentalPrint (2004, University of Granada, Department of Forensic Medicine and Forensic Odontology, Granada, Spain) that generates different comparison overlays from 3D dental cast images depending on the pressure of the bite or the distortion caused by victim-biter interaction. The procedure for generating comparison overlays is entirely automatic, thus avoiding observer bias. Moreover, the software presented here makes it impossible for third parties to manipulate or alter the 3D images, making DentalPrint suitable for bite mark analyses to be used in court proceedings.