Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 186
Ballistic Imaging 7 Three-Dimensional Measurement and Ballistic Imaging The striations on the edges of fired bullets and the textures impressed on the primer surface of a cartridge casing are, in their own right, images: representations of physical objects (imperfections and textures in a firearm’s barrel, breech face, and firing pin) depicted in a medium. These physical “images” are also inherently three-dimensional, produced by cutting, scraping, and etching. Part of the tension that has accompanied the use of photography in forensic firearms identification (see Section 3–F) arises from the fact that flat, two-dimensional representations of tactile, three-dimensional features is necessarily somewhat dissatisfying. Though it could be argued that any of the instantaneous views of bullet or cartridge case evidence through a comparison microscope is a two-dimensional perception, the ability to directly manipulate the exhibits—to alter their rotation and lighting—gives a three-dimensional experience that any single two-dimensional freeze-frame would lack. The basic objective of any ballistic image database is to collect some accurate representation of cartridge cases and bullets, derived so that entries can be compared and scored for similarity with others. The presence of an electronically coded representation of the physical objects obviates the need for direct access to the physical objects for comparison (though they would have to be directly examined for final confirmation). In theory, then, a three-dimensional model of a cartridge case or bullet—accurately conveying fine differences in depth but still capable of mathematical processing—would be ideally suited to the task. As advances have continued in the field of surface metrology in recent years, applications in the three-dimensional measurement of ballistics evi-
OCR for page 187
Ballistic Imaging dence have begun to emerge. As part of our charge to consider technical enhancements to the present National Integrated Ballistic Information Network (NIBIN) system—and to ballistic imaging, generally—consideration of three-dimensional measurement versus two-dimensional photography as the imaging standard was a natural pursuit. As Thompson (2006:10, 12) suggests, fully exploiting the three-dimensional aspects of the toolmarks left by firearms raises new levels of complexity relative to two-dimensional photography. “Striated [three-dimensional] toolmarks would be easy to match if, from the beginning to the end, they always stayed the same,” but they do not. Indeed, even fine striations—colloquially referred to as lines—do have a third dimension, depth, that can be appreciated “by using higher magnification”; ultimately, computer-based systems for analyzing striations will have to contend with the problem of deciding whether the different depths of “lines” convey any special significance. Moreover, Thompson (2006:12) notes: The dynamics of a bullet going down the barrel of a firearm, the downward movement of a fired cartridge case against the breech face of a Glock pistol, or the movement of a screwdriver across a door strike plate all leave 3-dimensional toolmarks that can change considerably in a short distance…. These features, toolmark angle, ammunition variability, [and] tool/barrel wear are features that an examiner considers during an examination and none of these can be [fully] captured in a [two-dimensional] photograph. In Chapter 8, we discuss experiments conducted on the committee’s behalf by the National Institute of Standards and Technology (NIST) using a prototype three-dimensional sensor on cartridge cases. This chapter provides basic background for that discussion, beginning with a discussion of the conceptual differences between two-dimensional and three-dimensional image acquisition technologies (Section 7–A). Previous efforts in three-dimensional measurement of ballistics evidence are described in Section 7–B, along with currently emerging three-dimensional products (7–C). 7–A ACQUISITION TECHNOLOGIES 7–A.1 Two-Dimensional Acquisition A two-dimensional approach to pattern comparison uses a photographic image of the object as the basic element. In considering the impact of two-dimensional imaging on the comparison process, there are several key factors—all driven by the fact that the image is a projection of light
OCR for page 188
Ballistic Imaging reflected off of three-dimensional objects onto a two-dimensional acquisition plane. These factors generally separate into geometry and photometry. The basic process of two-dimensional image formation involves several steps. Light rays are emitted from a source (or sources) located at specific geometric spots relative to the object and the sensor. Those rays follow standard optics as they emanate from the source to the object. When each ray strikes the object, it interacts with the surface of the object. There are several effects, depending on the material properties of the object. If the object is purely specular (a mirror), the ray is reflected back into the world at an orientation governed by the local geometry of the surface. More generally, however, the ray interacts with the surface. Some of the energy of the ray will be absorbed by the material, thus diminishing the total energy reflected back into the world. In addition, the microstructure of the surface will typically cause the ray to diffract, meaning that the amount of energy retransmitted off of the surface will vary as a function of the angle of emittance relative to the surface normal at the point. For example, in a purely matte surface, the amount of energy reflected from the surface varies as a cosine law. These effects are generally captured by the photometry of the situation, and techniques such as the bi-directional reflectance distribution function (BRDF) can be used to very accurately capture the reflectance properties of a material. Of course, this works for ideal materials, or materials whose properties can be measured in isolation. In more general settings, one uses approximations to capture the BRDF of a material. Once the light energy is reflected off of the surface, it obeys standard optics laws, and is captured by a sensor (camera). Here, the geometry of the situation will influence how many photons are captured at a single image element—the distance of the camera from the surface (typically not an issue in close-range imaging), the orientation of the cameras optics system and acquisition plane relative to the incoming rays, as well as other effects. In general, one can characterize several factors that influence the amount of light captured at a pixel (picture element) in a standard imaging device: position and strength of the light source; physical extent of the light source (assuming it is not a point source); use of multiple light sources; geometry relating light source positions to the surface material of the object being sensed; geometry of the object itself (see below); material properties of the object (this includes both changes in the material across the object, which will change the amount of light reflected independent of geometry (e.g., dirt or other defects may reduce the light and
OCR for page 189
Ballistic Imaging thus the appearance of a particular pixel) and the manner in which light, independent of the total incoming amount, is reflected from the surface); and geometry of the sensor relative to the object being sensed. Clearly, the overall orientation of a local patch of an object, both relative to the sources and to the sensor, is a major factor in determining how much light is reflected to the source. A naïve analysis, however, assumes that the object is spherical or cylindrical, that is, that all incoming rays strike the object and there is no occlusion. When the object has a more intricate surface, however, the situation gets more complex. This is especially true if there can be self-occlusion, that is, that some light rays may not reach part of the object because they are reflected by other parts of the object—casting self-shadows. Given all of these factors, one can see that there is a fundamental issue in comparing an image of a probe object against an image of a target object—one needs to ensure that the comparison of intensities in an image, or of some other feature extracted from the intensities, is actually reflecting the shape of the underlying object, and not some other factor. Many of these elements can be controlled. For example, using the same strength of light source, and fixing the position of the light source relative to the object will keep these factors constant across images. Normalization of the image intensities can also remove effects of the elements. Ensuring that the objects are cleaned in a consistent manner will remove material property changes from the images. Because of the shapes of bullets and shell casings and because their surfaces can be highly reflective (and hence high-glare), the geometry of the acquisition setup is very important. Keeping the orientation of the object the same with respect to the camera across acquisitions is very important. Given that one is primarily measuring striations on the surface of the object, self-shadowing and angle of reflectance effects are very critical, in order to ensure that the striations are both visible and have the same effect on intensities in the two images. One way to deal with this issue is to use multiple light sources—effectively to bathe the object in light from multiple directions. A standard technique is to use a ring of light sources surrounding the camera itself. This tends to reduce self-shadowing effects and reduces the impact of the specular reflection properties of metal objects. An alternative is to use multiple sources but to sequence them—that is, to take multiple images in the same geometry but illuminated from different directions. This can highlight striation patterns in one of the images that might get washed out in a bathing scenario. Other actions can be taken to reduce image variations not related to surface variations. In addition to controlling the lighting effects, the resolution of the image acquisition device, relative to the size of the object, is
OCR for page 190
Ballistic Imaging important. Since one is typically trying to capture image information about small markings on the object, there is a danger that those markings will get blurred out. Consider the geometry of the situation. While idealistically each light ray is reflected from a different infinitesimal patch of the surface, so that surface patches from the interior of a striation will reflect a different ray than a nearby unmarked surface patch, at some point all of those rays are captured by a patch of an acquisition device (leading to a pixel or picture element). If the pixels are small in comparison with the object size, then nearby rays—one from a striation, one from the nearby surface—will be captured by different pixels. However, if the pixels are too large, then these rays may project to and be integrated out by the same pixel. This blurring of the image can be crucial in this setting, so it is important to determine the size of standard striations and to ensure that the camera device is sufficiently high resolution to capture these changes in surface shape. 7–A.2 Three-Dimensional Acquisition Since the goal is to compare physical shapes of specimens—the probe object against a stored target object—an alternative is to try to directly measure the three-dimensional shapes. In other words, rather than trying to control or factor out all of the components that affect the image of an object, an alternative is to directly measure the shape. If one can do this, then the comparison can take place on shapes, rather than on image appearance of shapes, and all of the light interaction issues no longer matter. Three-dimensional surface measurement techniques have developed to include both contact and noncontact methodologies. A contact probe, such as a stylus, can directly measure the three-dimensional position of a point on a surface relative to a fixed coordinate frame. Repeated passes of such a contact probe can be used to more fully reconstruct a three-dimensional shape, and these direct surface measurements can be directly compared against other exemplars. In the particular context of ballistics evidence analysis, however, contact methods are problematic for several reasons. One problem is the size of the object being studied—a bullet or a casing. Most contact probes do not have the level of resolution necessary to build a sufficiently detailed three-dimensional reconstruction. As described more fully in the next section, a more fundamental difficulty is the potential for the evidence bullet or casing to be scratched or otherwise damaged using contact methods, potentially jeopardizing the chain of evidence. Noncontact methodologies have emerged that do have high resolution, certainly sufficient for the task of working with bullet or casing evidence. These noncontact methodologies include confocal microscopy, interferometry, or laser scanners. Each of these methods can be used to capture the three-dimensional shape of an object, without being subject to nonlinear intensity
OCR for page 191
Ballistic Imaging effects due to light reflection properties, or to self-shadowing effects. The main advantage, therefore, is being able to capture and directly compare three-dimensional shape information. However, the use of these highly sensitive methods can incur some disadvantages, chief among them: Cost—such sensors are usually much more expensive than optical camera systems. Speed—such sensors are typically much slower, taking orders of minutes or tens of minutes to acquire a three-dimensional image, rather than a fraction of a second. Noise—it is important to characterize the noise in the acquired measurements. If the noise is on the order of the depth of the striations, this will render the approach ineffective. In Chapter 8 we consider the performance of confocal microscopy, which operates on the principle of focusing a point of light on parts of a surface separately and measuring the intensity of returned rays of light rather than a pure reflectivity approach of illuminating the whole surface at once. In particular, light is concentrated through a pinhole aperture to reach the surface, and reflected rays pass through a second pinhole in order to filter out rays that are not directly from the focal point. A three-dimensional reconstruction can be built by varying the height of the pinhole apparatus, thus creating a series of thin two-dimensional slices from which three-dimensional heights can be derived by considering the vertical level at which the maximum level of light was reflected back from a particular point (Semwogerere and Weeks, 2005). The particular microscope tested in Chapter 8 makes use of a Nipkow disk, a spinning disk consisting of multiple pinholes, in order to collect information more rapidly from a wider lateral surface. 7–B PAST EFFORTS IN THREE-DIMENSIONAL IMAGING OF BALLISTICS EVIDENCE Seeking a way to reinforce firearms identification with a quantifiable and objective basis, Davis (1958) developed an instrument he called a “striagraph.” The striagraph recorded measurements from a stylus riding around the circumference of a recovered bullet as the bullet was rotated, providing three-dimensional surface measurement of the path around the bullet. The striagraph was never developed commercially, in part because of two key limitations that continue to affect the use of stylus methods for forensic analysis today: The method was not applicable to deformed or fragmented bullets (as are not uncommon at crime scenes), and the direct contact of the stylus could scratch or mark the bullet, corrupting
OCR for page 192
Ballistic Imaging the evidence (Gardner, 1979). A similar stylus method, dubbed the Balid system, was described in a conference presentation and summarized in the second issue of the AFTE Newsletter in 1969. The destructive nature of stylus profiling methods was later demonstrated by Blackwell and Framan (1980), using scanning electron microscopy to illustrate the deformation caused by using a stylus-based “Profilcorder” to trace the circumference of several bullets. Though stylus methods are infeasible for forensic analysis, methods for profilometry—the generation of one-dimensional vectors of height information—of ballistics evidence were pursued by later researchers. De Kinder et al. (1998) and De Kinder and Bonfanti (1999) analyzed bullet striations using three-dimensional profilometry; their scope used a reflected laser as a sensor and was capable of measuring height differences to 1 μm. De Kinder et al. (1998:299) performed preliminary analysis on 9mm Para bullets (including bullets from unfired rounds), as well as those fired through a Fabrique Nationale High Power pistol and recovered using either a water tank or cotton wool. They concluded by noting that “we hope to reduce [the disadvantage of lengthy data capture times] by setting up a procedure to extract a feature vector. This will probably no longer necessitate us to record the whole surface of a bullet, but only a few circumferences to obtain a representative data set of the surface topology.” De Kinder and Bonfanti (1999) extended this work, taking 151 scans (0.05 cm apart) beginning approximately 1mm from the end of the bullet, thus giving a set of profiles along a 7.5mm patch. (The first 34 scans were later found to be relatively noninformative and were dropped from analysis.) Each circumference measurement was taken with an overlap to account for striations split by the initial starting point. Data capture time was 4–5 hours per bullet, “which will be reduced by optimising the definition of the feature vector” (De Kinder and Bonfanti, 1999:87). To compare bullets, they constructed a correlation matrix consisting of the correlations between feature vectors for each of the land impressions (the bullets they studied had six land engraved areas). They took the trace of the resulting matrix as a summary measure for that specific alignment between the bullets; the six traces that arise from reordering the matrix for different land impression alignments were collected. They then compared the maximum value (the presumptive best match) to the average of the traces for non-corresponding alignments. So, for one case involving two bullets from the same gun, they found that the “sum of the correlation coefficients for corresponding striation marks” was 64 percent “higher than the average value for non-corresponding match.” In the sole case where they compared bullets from two different guns, the same factor came to 11 percent. They concluded that, of six cases where two bullets from the same gun were
OCR for page 193
Ballistic Imaging compared, “a well founded positive answer can be provided for about one in four cases, while for the two other comparisons, no clear answers can be given” (De Kinder and Bonfanti, 1999:92). In correlating the series acquired at different heights from the bullet base, they found that, “contrary to our expectations, optimal results (correlation coefficients larger than 80 percent) were not obtained for the scans closest to the back of the bullet, but for lines 80 to 100, corresponding to a distance to the base of about 2mm” (De Kinder and Bonfanti, 1999:89–90). Also focused on the problem of analyzing bullet striations, Bachrach (2002) developed SciClops to acquire profiles of bullet striations. This platform used confocal microscopy to derive a linear, topographic trace around the circumference of a bullet. This work would ultimately be extended to include analysis of a rich set of test-fired bullets, using gun barrels from nine different manufacturers and including more than 200 firings through each (the barrels were cleaned at one point in the firing, to determine the effect of that action on observed striations). The research suggested a three-way gradation in terms of the propensity of manufactured barrels to leave detectable and reproducible marks. A middle range of barrels and manufacturers worked best for toolmark deposition. At one extreme were relatively cheap firearms barrels whose less precise manufacturing standards added randomness to the observed markings and precluded easy matching; at the other were extremely high-end barrels that were so finely polished and machined as to render toolmarks too subtle to readily distinguish. Banno et al. (2004) acquired images from bullets (two from a Tanfoglio GT27 automatic pistol and another from a Browning model 1910 7.65mm) using a Lasertec HD100D-A confocal microscope. This microscope is capable of measuring a 3.2mm × 3.2mm patch with 450 × 450 pixels, with 0.02 μm height resolution. Images of land engraved areas were generated by connecting a 4 × 4 set of separate patches. Software aligned the different three-dimensional renderings, and similarity was assessed by differencing the aligned images and visualizing the results, shaded to indicate whether differences were within a 0.015mm tolerance. Images for the bullets fired from the same Tanfoglio pistol showed generally strong similarity, with higher differences generated when comparing features from the Tanfoglio and Browning test fires. Banno et al. (2004:240) illustrate but do not extensively analyze use of this measurement for other surfaces, including cartridge case markings. “This algorithm did well” in comparing firing pin impressions for two cartridges from the same weapon; though there is difference in texture along the wall of the firing pin impression, the hollows of the interior of the marks overlap almost exactly. Zographos et al. (1997) and Evans et al. (2004) advanced the “Linescan” system, a revised methodology for obtaining a composite two-dimensional
OCR for page 194
Ballistic Imaging image around the edge of cylindrical shape, such as a cartridge case or bullet. The system acquires images in a small window while the object is turned, resulting in a continuous imaging process rather than a stitch-together of related images. 7–C EMERGING PLATFORMS FOR THREE-DIMENSIONAL IMAGING OF BALLISTICS EVIDENCE In the past few years, Forensic Technology WAI, Inc. (FTI) has developed a bullet-only three-dimensional imaging system, dubbed BulletTRAX-3D.1 This new system “acquires two-dimensional and three-dimensional data in digital form from the entire bearing surface of a fired bullet to obtain its digital ‘signature’, specifically, a map of its surface topography” for a band around the bullet (Dillon, 2005:5). This differs from the standard Integrated Ballistics Identification System (IBIS) entry that requires operators to specify and image the separate land engraved areas. Graphically, this image data can be rendered onscreen in layers and, notably, as a planar surface that can be rotated and lit (altering both direction and type of simulated lighting) to see striations in relief. A software module also attempts to detect and display bands of consecutive matching striations, an emerging standard for quantifying bullet comparisons (see Section 3–B.3). Like its two-dimensional counterpart in IBIS, the comparison algorithm utilized by BulletTRAX-3D is proprietary information.2 As such, it is unknown how it differs from the standard two-dimensional IBIS in its comparison routines. However, a reading of Roberge and Beauchamp’s (2006) analysis, described below, suggests that the types of scores returned by BulletTRAX-3D are similar to those returned by IBIS. Roberge and Beauchamp (2006) report success in using FTI’s BulletTRAX-3D platform in a complicated test of bullet matching, making use of a set of 10 consecutively manufactured Hi-Point barrels. These button-rifled barrels are known to create major problems for direct visual comparison (see Section 2–D.1). Four bullets were fired through each barrel, and these were grouped into pairs; the objective was to match one 1 The exact rendering of the name of the system varies. The promotional brochure for the system uses a logo that depicts the “3D” part of the name in superscript—BULLETTRAX3D—but describes the system in text as BULLETTRAX-3D. However, Dillon (2005) and Roberge and Beauchamp (2006) used mixed case, calling it BulletTRAX-3D. 2 Dillon (2005:15) makes the remarkable statement that “the search algorithms employed … are proprietary in nature and not of direct interest to the firearms examiner.” Dillon suggests that “the examiner is less concerned with the search algorithms and much more concerned with the bottom line represented by the system’s list of high probability associations with other cases,” though how one can be confident in the “high probability” of suggested associations without any understanding of the algorithm’s process is not specified.
OCR for page 195
Ballistic Imaging group of 10 pairs (labeled 1–10) to the second (labeled A–K; an 11th pair with a different number of land impressions was inserted in this group). Roberge and Beauchamp (2006) exploited the pairwise nature of the test samples to create a training set of known matches; this gave them a sense of optimal “Max Phase” scores (see Chapter 4) to use as a decision rule and assign matches. Following the training phase, the testing was performed stage-wise—performing a set of comparisons, applying decision rules to pick out matches, removing those elements from the dataset, and repeating—until all assignments were made. Though a caption in Dillon (2005:10) touted BulletTRAX-3D (and its companion MatchPoint Plus display stations) as “the latest configuration of IBIS”—suggesting a replacement of IBIS—the system was originally positioned as a counterpart to IBIS. However, FTI has recently indicated a shift of its product line to focus on three-dimensional platforms, shifting the two-dimensional system currently deployed as the base for the NIBIN-system as the “IBIS Heritage” branch (see Box 4-1). Promotional materials for the three-dimensional systems emphasize that the three-dimensional systems are backward-compatible with the older two-dimensional systems; photographs are taken during the two-dimensional acquisition process and are offered as a layer that can be viewed onscreen in the three-dimensional system, so that photographs can presumably be subjected to the existing two-dimensional comparison process. It is unknown what changes have been made to account for three-dimensional measurement information in generating comparison scores in these new systems.
OCR for page 196
Ballistic Imaging This page intentionally left blank.