The use of machine learning for the determination of a type/model of firearms by the characteristics on cartridge cases.

Cartridge cases are commonly collected at crime scenes involving firearms. One of the stages in forensic examination is the determination of the type and model of firearms based on the class characteristics of these cartridge cases. A firearm examiner evaluates the class characteristics on the basis of their knowledge and experience, and by referring to collections of cartridge cases representing class characteristics of different firearms, special databases and reference books. However, this process is highly subjective. The novelty of this research is in developing objective methods of firearms determination by applying a machine learning approach. In this study, several Convolutional Neural Networks from Keras programming package were trained to determine the type/model of a firearm based on the class characteristics observed on cartridge cases from seven different categories of firearms. The prediction accuracies received by this method range from 71 to 81 percent for models based on different Convolutional Neural Networks, while using an ensemble of the machine learning models increased the accuracy to 88 %. The research demonstrates the efficacy of machine learning in enhancing accuracy and reducing subjectivity in firearm identification, highlighting its significant potential in forensic science applications.

Form follows function: Applying photographic content analysis to forensic firearm identification.

Drawing forensic conclusions from an image or a video is known as "photographic content analysis." It involves the analysis of an image, as well as objects, actions, and events depicted in images or video. In recent years, photographic depictions of objects suspected as illegal firearms have substantially increased, appearing on CCTV surveillance footage, captured by mobile phones and shared on social media. However, the law in Israel states that a person can be charged with illegally possessing a firearm only if it can be proven that the object is capable of shooting with lethal bullet energy. This becomes more challenging in cases where the firearm was not physically seized, and the evidence exclusively consists of images and video. In this study, photographic content analysis was applied to images and video where objects suspected as commercial or improvised firearms had been depicted. An image and event sequence reconstruction video databases of both firearms and replicas were created in order to better define firearm-specific functional morphological features. We demonstrate that it is possible to classify an object as a firearm by analyzing the functional, and not only the esthetic, morphology in images and video. It is also shown that event sequence reconstruction in video may be used to infer that an object suspected as a firearm has the capacity to shoot by confirming the occurrence of a shooting act or shooting process. Thus, photographic content analysis may be used to forensically establish that an object depicted in an image or a video is a firearm by ruling out other known scenarios, and without physically seizing it.

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.

Investigation of the variability in the tool marks produced by modified blank firing pistols on fired 7.65 mm projectiles.

Firearms identification has an important place in forensic ballistic investigations since the weapons are widely used in criminal offences. Firearm examiners resolve many case files, through the use of automatic systems or comparison microscopes. Advanced forensic technologies like BALISTIKA helps to record and analyze non standard ballistic evidence. In today's world, with the ease of access to materials and production technique information, pistols modified from blank firers are frequently encountered as crime tools. In this study, the characteristics of 7.65 mm fired cartridge cases obtained by controlled shots from blank firing modified pistols were examined, and their detection performances were compared by means of the Balistika system. Although distinctive differences are not expected after successive test shootings, balistically important changes were seen after the use of blank firing modified pistols and the 3D imaging system proved to be useful in observing such differences. The analyses showed that the modifications in weapons lead to variation in the ballistic characteristics and reduce the accuracy of the detection performance, which may result in flawed forensic decisions. It was also found that the deviations in ballistic impressions of modified blank firing pistols were greater than that of standard fabricated and hand-made pistols. This unique study contributed to the forensic sciences literature by focusing on the impact of modified weapons on ballistic characteristics.

Identification of bullets fired from air guns using machine and deep learning methods.

Ballistics (the linkage of bullets and cartridge cases to weapons) is a common type of evidence encountered in criminal cases around the world. The interest lies in determining whether two bullets were fired using the same firearm. This paper proposes an automated method to classify bullets from surface topography and Land Engraved Area (LEA) images of the fired pellets using machine and deep learning methods. The curvature of the surface topography was removed using loess fit and features were extracted using Empirical Mode Decomposition (EMD) followed by various entropy measures. The informative features were identified using minimum Redundancy Maximum Relevance (mRMR), finally the classification was performed using Support Vector Machines (SVM), Decision Tree (DT) and Random Forest (RF) classifiers. The results revealed a good predictive performance. In addition, the deep learning model DenseNet121 was used to classify the LEA images. DenseNet121 provided a higher predictive performance than SVM, DT and RF classifiers. Moreover, the Grad-CAM technique was used to visualise the discriminative regions in the LEA images. These results suggest that the proposed deep learning method can be used to expedite the linkage of projectiles to firearms and assist in ballistic examinations. In this work, the bullets that were compared were air pellets fired from both air rifles and a high velocity air pistol. Air guns were used to collect the data because they were more accessible than other firearms and could be used as a proxy, delivering comparable LEAs. The methods developed here can be used as a proof-of-concept and are easily expandable to bullet and cartridge case identification from any weapon.

Examination of the possibility to use Siamese networks for the comparison of firing pin marks.

One of the most discussed issues in forensic firearms identification is the subjectivity of conclusions. The main part of firearms examiners' work is to make a microscopic comparison of the marks on cartridge cases and bullets. In this process, examiners have to decide if the quantity and the quality of the observed characteristics are sufficient for identification. This decision is based on the personal experience of an examiner, so examiners with different backgrounds can come to different conclusions, and this fact presents a problem. Besides, the calculation of the error rate for this type of examination is a debatable issue. Different mathematical and statistical models were proposed, and computer-based algorithms were developed in order to avoid subjectivity and to determine error rates. This article investigates the possibility to use methods of machine learning for the comparison of marks of the firing pin impressions on cartridge cases. In the research, the Siamese network model, which included two similar Convolutional Neural Networks, was prepared and trained. For the training and validation of the model, the database of firing pin impressions was prepared. This database included images of cartridge cases discharged from 300 firearms that came from regular casework and clone images used for data augmentation. The model was trained and examined using the validation part of the database. The metrics, such as accuracy, sensitivity, and specificity were calculated. The results of the research show the possibility of using the Siamese network for building an objective forensic firearms examination system with a known error rate.

Objective evaluation of similarity scores derived by Evofinder® system for marks on bullets fired from Chinese Norinco QSZ-92 pistols.

In recent years, many studies have been conducted in the field of firearm identification with the objective of providing an objective method of evaluating the comparison of cartridge cases. However, less attention has been paid to the objective evaluation of bullet comparisons. In this study, 1 000 registered Chinese Norinco QSZ-92 pistols were used, and a database of 2 996 bullets was constructed. Both the receiver operating characteristic (ROC) curve and the score-based likelihood ratio method were used to objectively evaluate the similarity scores derived by the Evofinder® system. The results indicate that this system has excellent ability to discriminate between the selected pistols. This paper proposes an objective evaluation method, which serves as a response to the ongoing debates regarding the foundation of the discipline.

Potential impacts of 3D modeling and 3D printing in firearm toolmark examinations.

In this work, two gun barrels were 3D-printed using a metal alloy material from the same 3D digital model created in silico. After assembling the barrels in a reference Colt 45 Auto semi-automatic pistol, a total of 100 rounds of ammunition were successfully test-fired through each barrel. Heavy and gross striations were observed on all fired bullets. The striations on bullets discharged from the corresponding 3D-printed gun barrel were found to be identifiable. Moreover, bullets fired from one 3D-printed gun barrel were easily excluded from those fired from the other 3D-printed gun barrel by visual examination under a comparison microscope. The resulting unconventional striations that were observed were apparent to an experienced firearm and toolmark examiner. These features could provide valuable investigative leads related to the 3D printing process. Since 3D printing has become an option for firearm manufacturing, the forensic science community should establish a knowledge base associated with the toolmark features generated by the 3D-printed products.

Quantitative characterisation of ballistic cartridge cases from micro-CT.

Evaluation of cartridge cases is essential within forensic ballistic analysis and is used in an attempt to establish a connection to the weapon used to fire it. This study consists of two experiments. The aims of Experiment 1 were to establish whether micro-CT is appropriate and repeatable for ballistic cartridge case analysis and if measurements can be extracted repeatably and reliably. Experiment 2 aimed to compare cartridge cases from two weapons to establish the magnitude of variation within and between weapons. A total of 48 cartridge cases fired by two distinct weapons were collected and micro-CT scanned to a high resolution. One randomly selected cartridge was scanned ten times under the same conditions to ensure repeatability of the scanning conditions in Experiment 1. Three novel measurements to quantitatively assess the firing pin impressions were proposed in Experiment 1 and comparatively analysed from two weapons in Experiment 2. Experiment 1 showed that micro-CT is an effective and highly repeatable and reliable method for 3-dimensional imaging and measurement of ballistic cartridge cases. Furthermore, high agreement for inter-rater reliability was found between five raters. Quantitative micro-CT analysis of the firing pin impression measurements in Experiment 2 showed a significant difference between the two studied weapons using Welch's t-test (p < 0.01). This study shows the advantage and reliability of utilising micro-CT for firing pin impression analysis. Quantitation of the firing pin impression allows distinction between the weapons studied. With expansion to further weapons, application of this methodology could complement current analysis techniques through classification models.

Reproducibility and persistence of individual marks on 3000 fired bullets from five Chinese Norinco QSZ-92 9 × 19 mm pistols, and the search performance of Evofinder® to a 1000 firearm database of the same model firearm.

For firearm identification, foundational validity based on the reproducibility and persistence of characteristic marks must be established. We investigate the fired bullets of five Chinese Norinco QSZ-92 9 × 19 mm pistols over 3000 shots. The first 50 fired bullets are recovered, whereas every 50th fired bullet is recovered from the 51st to 3000th round. As such, 109 bullets are available for each pistol, and totally 545 bullets are introduced into the Evofinder® system. A large background database comprising 3000 bullets fired from 1000 registered QSZ92 9 × 19 mm pistols is used as interference. Both on-screen analysis and automatic comparison are performed. The first fired bullets from the five pistols are separately correlated with the database. The results show that although the similarity for known match bullets changes slightly as the shot number increases, the land-engraved area (LEA), groove-engraved area (GEA), and slippage marks can be reproducibly transferred to the fired bullets in consecutive shots. The Evofinder system ranks all known match bullets on the top of the correlation result with the combination of LEA, GEA, and slippage marks.

Sharing ballistic data across Europe: A prototype network between France and Switzerland using Evofinder®.

Today, there is a real political urge to see the sharing of ballistic data intensify across Europe mostly due to recent events such as terrorist attacks. However, technical constraints remain and two main options are being discussed. The first one relies on a centralized common database, implying a vendor monopoly for all Europe and a unified protocol among member states. The second one advocates for a distributed framework relying on existing national infrastructures and leaving each country responsible for its own protocols. This article describes a prototype network linking Switzerland and France using the Evofinder® system by ScannBI. We will first focus on how this network was set up, and then report some results from tests conducted to assess the viability of the concept. These results demonstrate that the second option cannot be discarded and pave the way for a distributed network. This solution appears to be cheaper, more adaptable and answers the practical needs of member states.

Beretta barrel fired bullet validation study.

A report published in 2016 by the President's Council of Advisors on Science and Technology (PCAST) criticized studies that have been published regarding the discipline of firearm identification. This study was designed to answer some of these criticisms and involved 30 consecutively manufactured Beretta brand 9 mm Luger caliber barrels. This study had an "open set" design to help the discipline of firearm identification establish "Foundational Validity" which is outlined in the PCAST report. Seventy-two qualified firearm examiners completed and submitted answers for this study that included 15 knowns and 20 unknowns. There were an additional 5 firearms with similar characteristics as the Beretta barrels that were also included as unknowns which provided "known non-match" comparisons. Test sets were created using the random function in Microsoft Excel. Collaborative Testing Services (CTS) funded, facilitated, distributed the tests, and collected the answers from qualified firearm examiners throughout the United States and the world. Firearm examiners were able to complete the test of fired bullets with a low error rate. The error rate for the corrected data was 0.08% (1 in 1250) with the lower confidence interval as low as 0.01% (1 in 10,000) and the upper confidence interval being as high as 0.4% (1 in 250).

Evaluating Likelihood Ratio (LR) for firearm evidence identifications in forensic science based on the Congruent Matching Cells (CMC) method.

Firearm evidence identification has been challenged by the 2008 and 2009 National Research Council (NRC) reports and by legal proceedings on its fundamental assumptions, its procedure involving subjective interpretations, and the lack of a statistical foundation for evaluation of error rates or other measures for the weight of evidence. To address these challenges, researchers of the National Institute of Standards and Technology (NIST) recently developed a Congruent Matching Cells (CMC) method for automatic and objective firearm evidence identification and quantitative error rate evaluation. Based on the CMC method, a likelihood ratio (LR) procedure is proposed in this paper aiming to provide a scientific basis for firearm evidence identification and a method for evaluation of the weight of evidence. The initial LR evaluations using two sets of 9mm cartridge cases' breech face impression images with different sample sizes, imaging methods and ammunition showed that for all the declared identifications of the tested 2D and 3D image pairs, the evaluated LRs for the least favorable scenario were well above an order of 106, which provides Extremely Strong Support for a prosecution proposition (e.g. a same-source proposition) in a Bayesian frame. The LR evaluations also showed that for all the declared exclusions of the tested 3D image pairs, the evaluated LRs for the least favorable scenario were above an order of 102, which provides Moderately Strong Support for a defense proposition (e.g. a different-source proposition) in a Bayesian frame.

Firing pin impressions: a valuable feature for determining the orientation of the weapon at the time of shooting.

Sometimes the firearm forensic examiner is required to provide information useful to discriminate between suicide, homicide or accident, or between contradictory reconstructions of the events (like attempted murder versus accidental discharge). In such situations, knowledge of the position and orientation of the firearm at the time of firing can be of fundamental help for the reconstruction of events. To achieve these goals, the analysis of the firing impressions is very important. In this study, the cartridge cases shot with three different revolvers aiming at three different spatial orientations (vertical upwards, horizontal or vertical downwards) were studied. The depth and morphology of the firing pin impression was characterised by optical microscopy and quantified by a surface topography analysis. The orientation of the firearm significantly modified the morphology and depth of the firing pin impression: ammunition fired upwards had the deepest firing pin impression, those fired downwards had the shallowest. Such behaviour was attributed to the different geometry of the firing pin-primer cup interaction and to the different pressure exerted by the primer as the orientation of the weapon changes. Therefore, this work has shown that a suitable protocol of morphological and topographical analysis can be set up to understand if a shot by a revolver was fired holding the weapon upwards, downwards or horizontally.

Three-Dimensional Analysis of Cartridge Case Double-Casts.

Due to the shot-to-shot variability in tool mark reproduction on fired cartridge cases, a method of replication is needed for the creation of training and testing sets. Double-casting is one method that has been used for this application, but the accuracy and variability of this method needs to be characterized. Three firearms were used to fire 25 cartridges each to create the master cartridge cases. The double-casting method consists of creating a silicone mold of the master cartridge case. A plastic resin mix is then poured into the mold to create the double-cast reproduction. Fifteen double-casts of each of the 75 fired cartridge cases were created across different silicone molds to analyze within- and between-mold variability. The master cartridge cases and double-casts were scanned with a confocal microscope (Sensofar® S neox) to create three-dimensional representations of the surfaces. Two similarity metrics were used for the objective comparison of the double-casts to their master cartridge cases: the areal correlation coefficient (ACCFMAX ) and the number of congruent matching cells (CMC). The ACCFMAX and CMC data, along with visual examinations, showed that the double-casting method produces accurate reproductions. Within-mold variability was found to be minimal, and between-mold variability was low. These results illustrate that double-casting can be applied for training and testing purposes.

Fired bullet signature correlation using the Congruent Matching Profile Segments (CMPS) method.

We introduce the Congruent Matching Profile Segments (CMPS) method for objective comparison of striated tool marks and apply it to bullet signature correlations. The method is derived from the congruent matching cell (CMC) method developed for the comparison of impressed tool marks. The proposed method is designed to increase comparison accuracy by addressing the comparison challenges caused by striae profiles with different lateral scales, varying vertical (height) scales, and sections that are poorly marked or have little to no similarity. Instead of correlating the entire profiles extracted from striated tool marks, the method divides one of the compared profiles into segments. Each segment is then correlated with the other profile. The CMPS method uses the normalized cross-correlation function with multiple correlation peak inspection to determine the number of profile segments that have both significant topography similarity and a congruent registration position. Initial tests were performed on the land engraved areas (LEAs) of 35 bullets fired from 10 consecutively manufactured pistol barrels. The results show clear separation between the CMPS scores of the 549 known non-matching (KNM) LEA profiles and the 46 known matching (KM) LEA profiles. These results are an improvement over those obtained using the correlation coefficient score of whole profiles. The large number of CMPS segment correlations may facilitate a statistical approach to error rate estimations.

A Multimodal Fusion Approach for Bullet Identification Systems.

In the field of forensic science, bullet identification is based on the fact that firing the cartridge from a barrel leaves exclusive microscopic striation on the fired bullets as the fingerprint of the firearm. The bullet identification methods are categorized in 2-D and 3-D based on their image acquisition techniques. In this study, we focus on 2-D optical images using a multimodal technique and propose several distinct methods as its modalities. The proposed method uses a multimodal rule-based linear weighted fusion approach which combines the semantic level decisions from different modalities with a linear technique that its optimized modalities weights have been identified by the genetic algorithm. The proposed approach was applied on a dataset, which includes 180 2-D bullet images fired from 90 different AK-47 barrels. The experimentations showed that our approach attained better results compared to common methods in the field of bullet identification.

Pilot study of feature-based algorithm for breech face comparison.

A novel feature-based method, which is scale invariant feature transform (SIFT) and RANdom SAmple Consensus (RANSAC) integration algorithm, is introduced to promote the automated identification of the breech face impression, the most common mark left on the cartridge used for firearm evidence. SIFT algorithm is employed to extract the local extrema from examined impression as keypoints representing its invariant features, and to build the feature descriptor for each keypoint based on its local gradients in neighborhood. RANSAC is used to improve the matching performance among these keypoints and feature descriptors. With hypothesize-and-verify methods, RANSAC is able to construct the best model fitting initial matching pairs of keypoints and to guarantee the robust comparison result. Validation tests using 40 cartridge cases fired from pistols with 10 consecutively manufactured slides yielded a clear separation result, which strongly supports the effectiveness of the ensemble algorithm of SIFT and RANSAC. This application indicates the practical feasibility of feature-based algorithm and image processing technique in forensic science.

A simple method to compare firing pin marks using stereomicroscope and Microsoft office (Windows 8) tools.

Pertinent marks of fired cartridge cases such as firing pin, breech face, extractor, ejector, etc. are used for firearm identification. A non-standard semiautomatic pistol and four .22rim fire cartridges (head stamp KF) is used for known source comparison study. Two test fired cartridge cases are examined under stereomicroscope. The characteristic marks are captured by digital camera and comparative analysis of striation marks is done by using different tools available in the Microsoft word (Windows 8) of a computer system. The similarities of striation marks thus obtained are highly convincing to identify the firearm. In this paper, an effort has been made to study and compare the striation marks of two fired cartridge cases using stereomicroscope, digital camera and computer system. Comparison microscope is not used in this study. The method described in this study is simple, cost effective, transport to field study and can be equipped in a crime scene vehicle to facilitate immediate on spot examination. The findings may be highly helpful to the forensic community, law enforcement agencies and students.

Objective Evaluation of Subclass Characteristics on Breech Face Marks.

Subclass characteristics can be found on the breech face marks left on spent cartridge cases. Even if they are assumed to be rare and their reported number is small, they can potentially lead to false associations. Subclass characteristics have been studied empirically allowing examiners to recognize them and to understand in which conditions they are produced. Until now, however, their influence on the identification process has not been studied from a probabilistic point of view. In this study, we aim at measuring the effect of these features on the strength of association derived from examinations involving subclass characteristics. The study takes advantage of a 3D automatic comparison system allowing the calculation of likelihood ratios (LRs). The similarities between cartridge case specimens fired by thirteen S&W .40S&W Sigma pistols are quantified, and their respective LRs are computed. The results show that the influence of subclass characteristics on the LRs is limited, even when these features are prevalent among the potential sources considered in a case. We show that the proportion of firearms sharing subclass characteristics should be larger than 40% of the pool of potential firearms for the effect to be significant.