Citation
Meijerman, L. (2006, February 15). Inter- and intra-individual variation in earprints. Barge's Anthropologica, Leiden. Retrieved from https://hdl.handle.net/1887/4292
Version: Not Applicable (or Unknown)
License: Licence agreement concerning inclusion of doctoral thesis in theInstitutional Repository of the University of Leiden
Downloaded from: https://hdl.handle.net/1887/4292
INDIVIDUALIZATION OF EARPRINTS: VARIATION IN PRINTS OF
M ONOZYGOTIC TW INS
L.M eijerman,A.Thean,C.van der Lugt,R.J.van M unster,G.van Antwerpen and G.J.R.M aat
Abstract
9.1 Introduction
The Bertillon identification system was used to identify recidivists by law enforcement authorities in the United States and parts of Europe during the early nineteen hundreds. Among other characteristics, eleven measurements of the body were registered. It was assumed that no two people would have all measurements exactly the same. In 1903, the system appeared not accurate enough to tell the difference between W ill and W illiam W est, two similar prisoners of the U.S. Penitentiary at Leavenworth, Kansas. In contrast, a fingerprint comparison supposedly did distinguish them as two different people. Although the two prisoners denied it themselves, it was later discovered that they were identical twin brothers (Scottish Criminal Record Office, 2002). This inability to classify twin brothers as separate individuals was said to have marked the end of reliance on the Bertillon system. According to Cole (2004), fingerprint advocates concocted this story as an appealing creation myth. The anthropometric measurements of the two prisoners did not match, nor were they already fingerprinted in 1903. True, or otherwise, this story highlights a desirable property of biometric measurements, namely that they should be able to distinguish between all individuals, even between monozygotic (identical) twins.
Although the ears of monozygotic twins may show strong similarities, certain differences have been reported in the literature. The presence and size of an auricular tubercle may vary among twin members (Quelprud, 1936), as may the degree of folding of the helix (Geyer, 1936) and the shape and length of the earlobe (Quelprud, 1935). The length and width of the external ear may further vary between identical twin members. Two independent studies recorded an average difference of less than 1% for both ear length and width between identical twin members (Leicher, 1928; Von Verschuer, 1927). However, particularly ear length estimated from different earprints of a single ear appears to be subject to variation depending on the amount of pressure that was applied to the surface during the process of listening (Neubert, 1985; Saddler, 1996). But striking differences in imprinted ear width have also been observed (Meijerman et al., 2005b). W e therefore presume that a difference of less than 1% in either of these dimensions would make the variation between the ears smaller than the potential variation in multiple prints of a single ear29.
29
In this study we aim to investigate inter- and intra-individual variation in earprints of monozygotic twins in order to investigate their value for individualization. A frequently quoted – though oversimplified – definition for the process of individualization is provided by Tuthill (1994): “The individualization of an impression is established by finding agreement of corresponding individual characteristics of such number and significance as to preclude the possibility (or probability) of their having occurred by mere coincidence, and establishing that there are no differences that cannot be accounted for”. In order to assess the degree of similarity that may occur in prints of different ears, we first describe the corresponding characteristics in prints of different twin members. We further describe differences that 'we cannot account for' i.e., that in our opinion allow for the distinction between prints from the individual members. For this, we evaluate differences in both content (i.e., presence, shape and intensity of imprinted features) and geometry (i.e., position of imprinted features).
A further goal of our study is to explore two different methods that may be used to semi or fully automatically distinguish between prints from the individual members. These methods could be applied to the design of an automated classification or matching system for earprints. For the development of such an automated matching tool, and in order to establish the evidential value of earprints, it is of fundamental importance to ascertain the degree to which inter-individual variability in prints dominates that of intra-individual variability. By processing the earprints of monozygotic twins we purposely select pairs of individuals for whom the degree of inter-individual variation is low. Our assumption is that the properties of the atypical subpopulation of twins can be used to make inferences about the general population e.g., features or techniques that can be used to distinguish between the earprints of monozygotic twins are likely to be useful for the classification and individualization of earprints in general.
9.2 Data
We have collected right earprints from six pairs of monozygotic twins (three male and three female pairs) of various ages. Information on the sex and age of the twin pairs is shown in Table 9.1. Every participating twin member was asked to listen three times for sound at a surface, using the right ear. The resulting earprints were dusted with aluminium powder, and lifted using Black Gel Lifters.
Each of the resulting 36 prints was scanned at 600 dpi (2100*3000 pixel). Colours were inverted to achieve a black on white appearance. A selection of prints from each twin pair is illustrated in this study (Figs. 9.1a,b-9.6a,b); all included prints are available at www.tno.nl/earprint.
___________________________________________________________________________ Table. 9.1. Information on the twin pairs, on the number of annotations per print, and the resulting number of variables used for cluster analysis.
___________________________________________________________________ Annotations per print
Twin no. Gender Age
Pressure-areas Angles/notches
Variables for cluster analysis
1 Male 18 9 2 55 2 Male 41 6 3 36 3 Female 36 6 2 28 4 Male 32 6 2 28 5 Female 26 3 2 10 6 Female 52 4 3 21 9.3 Methods
9.3.1 Account of similarities and differences
___________________________________________________________________________
___________________________________________________________________________
___________________________________________________________________________ Fig. 9.1-6 continued. Numbers in panel a correspond to characteristics mentioned in section 3.1; an ‘x’ in panel b indicates a pressure-area used for the cluster analysis, and a ‘+’ the position of a notch or angle.
members were described. A difference in the position of features between these prints was emphasized by digitally overlaying a colour-inverted copy of a print on the print of its twin using Adobe Photoshop software. Instead of selecting the two representative prints for each twin pair at random, the pair of prints that appeared to show the greatest degree of similarity was chosen for this purpose. The aim was to make the task of discriminating between prints of the twin members as difficult as possible.
The digital overlays provide an overall indication of the degree of similarity between prints of different twin members (Figs. 9.1c-9.6c). Dark areas show where the contact points of the first print have no counterpart in the second print, while light areas show where contact points in the second print have no counterpart in the first print; areas common to both prints appear grey. A judgement was made as to whether the differences described allowed for the different twin members to be discriminated.
9.3.2 Semi-automated comparison of the position of imprinted features
The degree of variation in the geometrical position of features between two earprints is difficult to quantify by eye. We therefore wanted to determine whether classification methods based solely on differences in geometry would allow the prints of twin members to be distinguished. We chose to use only those features that may show potential for automated extraction. Angles or notches in the outline of imprinted features may be amongst such features, as may pressure-areas, i.e., those areas that appear relatively intensity-strong in a print. They may be detected automatically using pixel intensity, and a centroid (centre of gravity) may then be calculated. The pattern formed by these centroids could provide an objective means of classification. For this exploratory study, however, we chose to use the estimated geometrical centre of the pressure-areas.
this mathematical procedure one attempts to identify relatively homogeneous groups of objects, based on the selected variables, using an algorithm that starts with each object in a separate cluster and combines clusters until only one is left. Table 9.1 provides information on the number of annotations per print for this classification method and the resulting number of variables used for cluster analysis.
9.3.3 Fully automated comparison of earprints
We have applied a method to automatically match earprints. The fact that the method requires no human input confers two main advantages. Firstly, it allows objective and repeatable earprint comparisons that are free from observer bias. Secondly, it has the potential to handle tasks that are too time-consuming for a human expert to consider. The method consists of the following steps: image pre-processing, keypoint30 detection, keypoint matching and similarity metric definition. The method is illustrated in Figure 9.7 and described in more detail below.
In order to reduce processing time, the full-resolution 2100u3000 pixel (600 dpi) images were first resampled to smaller 420u600 pixel images using a Lanczos filter. Keypoint (salient region) detection was performed using the Difference of Gaussians (DoG) operator (Lowe, 2004a). Keypoint description was performed using the Scale Invariant Feature Transform (SIFT) algorithm of Lowe (2004a), which was used to transform the pixels values in each of the detected regions into a 128-dimensional feature vector. The SIFT algorithm describes an image patch using the distribution of the image gradient directions across the patch and is robust to small translations of the image patch. The resulting feature vector is scale and rotation invariant but a measure of the characteristic scale and orientation of each patch is nevertheless determined. For the current application the characteristic orientation of each patch is used to filter out bad matches by imposing an ad hoc limit of 30 degrees on the maximum acceptable difference in orientation between two patches. Within this limit the orientation difference between corresponding image patches is measured. This is useful for earprint alignment, a problem which has hampered alternative methods (Meijerman et al., 2004c): note that no alignment step is necessary for the current method owing to the rotation-
30
__________________________________________________________________________
___________________________________________________________________________ Figure 9.7 Keypoint matching illustrated.
For each pair of illustrations, the print on the left is the same. The upper pair shows all detected keypoints in two different prints of the same ear. For the middle pair dotted lines join matching keypoints. In the lower pair 3 keypoint matches to a print from a different ear are shown. (Note that prints do not need to be aligned).
invariant property of the feature vector. Both the DoG operator and the SIFT algorithm were implemented using the executable code provided by Lowe (2004b).
Keypoint matching is done in the following way for a pair of prints. Firstly, keypoints are found and described automatically for each print. The number of keypoints detected depends on the image content: the number of keypoints per print for the current sample of 36 earprints varied between 850 and 3081, the mean was 1614. Each keypoint in the query print is selected in turn and compared with all other keypoints in the other print (candidate print) in order to look for keypoint matches. The best keypoint match is defined as that for which the Euclidean distance in the SIFT feature space is minimum. The ratio between the Euclidean distance to the best match and the Euclidean distance to the second best match is used as a measure of the match likelihood; ratios close to zero indicate unusual matches, whereas ratios close to one indicate more ambiguous matches.
Next, keypoint matches are ranked according to their likelihood ratios and regions with low likelihood ratios are used to search for a geometrical transformation that maximizes the number of region matches; corresponding keypoint pairs in each image are used to define a geometrical model hypothesis (consisting of rigid rotation and translation only) which is discarded or accepted according to whether it improves the number of matches found between the images. After a geometric model is chosen, constraints on the minimum allowed keypoint separation (in this case 4 pixels) along with the keypoint orientation are used to filter out bad matches. We choose to define a similarity metric as the number of accepted keypoint matches found for a pair of images; for prints from unrelated individuals we expect few keypoint matches, whereas for two prints from the same individual we expect many good keypoint matches.
9.4 Results and discussion
9.4.1 Account of similarities and differences
Figures 9.1 to 9.6 provide examples of the studied prints. In each figure, panel a shows an earprint of one of the twin members, referred to as ‘individual a’ in the descriptive section of this text. Panel b shows a print from the other twin member (‘individual b’). Panel c shows a digital overlay of these two prints. Illustrated prints were thought to show the greatest degree of similarity among prints of different twin members. Although only one print is shown per individual, similarities and differences mentioned in the following section apply to all studied earprints of each twin pair or twin member, unless otherwise stated. For each pair, the numbered arrows in the illustrated print of ‘individual a’ correspond with the similarities and differences mentioned in this section.
An opinion on the diagnostic value of mentioned differences is expressed. Differences that are not considered diagnostic, i.e., that in our opinion do not allow for the exclusion of one twin member as a possible source of the print, may still be useful for individualization. A comparison of a greater number of prints from twin pairs will provide further insight into the stability of mentioned differences, potentially improving the likelihood for one of the two twin members as the source of a print.
It has to be noted that this descriptive part of our study is subject to observer bias. The examiner knew that the sample contained prints from monozygotic twins, and also knew which prints originated from which ear. The opinion on the diagnostic value of observed differences is further subject to subjectivity. The account of similarities and differences is, however, of value to those familiarising themselves with variation in prints. To evaluate the evidential value of earprints and the competence of examiners, a blind test is recommended.
Pair 1 (Figs. 9.1a-c)
Striking similarities in prints of these twin members are the angle in the inner margin of the superior section of the helix [1] and the position of a series of pressure-areas [2], among which the imprint of an auricular tubercle [2'].
are the more extended scapha [3], the very pronounced pressure-area in the anthelix [4], the more pronounced posterior section of the helix [5], the slightly less angulated inner margin of the anthelix [6], and the large notch at the lobe [7]. From experience we know that the imprinted surface of the helix, particularly the posterior section, may vary among prints of a single ear. The extent to which the scapha is continued as a low-intensity area or lacuna in the lobe may also vary to some extent. The same is assumed for the curvature of the anthelix and the intensity of its pressure-areas.
The greatest difference between prints of the two individuals is the large notch in the posterior margin of the lobe [7]. The posterior margin is well-demarcated in prints of individual a, with clear pressure points on both sides of the large notch. We would therefore not expect to see a gradually curved margin – as shown in prints of individual b – in prints from the same ear. We believe this difference provides a good diagnostic clue for individualization.
The difference in the position of imprinted features, shown in Figure 9.1c, is not judged to exceed the range of potential intra-individual variation based on our experience with earprints to date.
Pair 2 (Figs. 9.2a-c)
Some similarities are found in the inner margin of the superior section of the helix, particularly the small notch in the inner margin [1], although this notch was not very obvious in one of the three prints of individual b. The characteristic anterior section of the anthelix [2] appears relatively similar in all prints. The imprint of an auricular tubercle [3] is furthermore present in all prints, although more prominently in prints of individual a.
quality of the earprint.
The inter-individual variation in the position of imprinted features, shown in Figure 9.2c, does not appear to be outside the range of potential intra-individual variation.
Pair 3 (Figs. 9.3a-c)
Again, some similarities are found in the inner margin of the helix, which is somewhat angulated [1], although more apparently so in prints of individual a. Other similarities are the expansion of the superior part of the helix [2] that occurs in prints of both twin members (although this part of the helix was not visible in one print of individual b because hairs had covered the ear at this position during the listening process) and the discontinuation of the posterior section of the helix [3].
Dissimilarities are found in the width of the posterior section of the helix [4], the continuation of the intertragic notch [5], the intensity of the posterior section of the antitragus [6], the prominence of the crus of helix [7], the shape of the anthelix, particularly the curvature of the stem and the position of the anterior crus [8], and finally the distance between the superior part of the anthelix and the helix [9]. We think that particularly the last two differences provide clues to distinguish between prints of the members of this twin. The crus of helix being so prominently imprinted in prints of one twin member, yet practically absent in prints from the other could be an additional clue for individualization. However, depending on pressure distribution, such variation may occur in prints of a single ear.
Although the shape and position of the anthelix and the helix may vary to some extent in prints of the same ear, the inter-individual variation as visualized in Figure 9.3c appears rather large to us to be mistaken for intra-individual variation.
Pair 4 (Figs. 9.4a-c)
The base of the helix, however, is very faintly imprinted in prints of individual a [4], while this section (including some of the pre-auricular area) is more prominently imprinted in prints of individual b. Further differences are found in the posterior section of the helix that is less wide or even interrupted in prints of individual a. This section contains a pressure-area in prints of both members of the twin [5], while in prints of individual b the helix expands at this position. The area inferior to this expansion is further of relatively low intensity in prints of this individual, while it is of quite high intensity in prints of individual a [6]. In prints of the latter individual, the anterior and posterior margins of the anthelix appear less sharply angulated [7] and the outline of the intertragic notch is not imprinted [8].
Particularly the angle of the anthelix may prove to be very useful for the process of individualization. When comparing prints other than those depicted in Figure 9.4, the differences between the two twin members become more obvious, as a larger surface of the anthelix becomes imprinted in prints of subject b. We feel that particularly the difference in the posterior margin of this feature could be considered diagnostic, providing the anthelix is sufficiently imprinted. Variation in applied force or the duration of listening may, however, influence width and length of imprinted features to some extent. For some ears, small changes in force may have a relatively large effect on the print appearance (Meijerman et al., 2004c). As the superior section of the anthelix is relatively narrow in most of the studied prints, we assume that this section could on occasion be absent, obscuring the curvature of the anthelix.
Pair 5 (Figs. 9.5a-c)
is more convexly curved in prints of individual a [5]. The inferior section of the helix is further absent [6], while it is prominently imprinted in prints of her sibling, and, when visible, the pre-auricular crease [7] is situated at greater distance from the tragus.
Apart from the position of tragus and antitragus pressure-areas, that we consider too great to be mistaken for intra-individual variation, we believe the imprints of the anthelix and the helix to be diagnostic when discriminating between prints of the twin members.
Pair 6 (Figs. 9.6 a-c)
Similarities are found particularly in the inferior sections of the prints of this pair. In prints of both members, the anthelix is situated closely to the helix and only the stem of the anthelix is imprinted [1]. The posterior auricular furrow leaves a similar lacuna between anthelix and antitragus [2] and the helix is slightly notched at the transition to lobe [3].
Fig. 9.6c shows that the imprint of the helix is so different that, in our opinion, there would be no doubt as to which print is left by which ear. In all prints, there is a pressure-area in the posterior section of the helix. In prints of individual a, this pressure-area is quite apparent [4] and the helix is interrupted on both sides of this pressure-area. In prints of her sibling, there are up to three pressure-areas present in the posterior section of the helix. From Fig. 9.6c it appears that the most inferior pressure-area of this series would approximately correspond to a void in the helix of individual a [5]. As this section of the helix is shaped so differently in prints of both twin members, the position of potentially corresponding characteristics is, however, difficult to interpret.
9.4.2 Semi-automated comparison of the position of imprinted features
Table 9.1).
The results of the cluster analyses using these parameters are encouraging. In each analysis, the three prints of a single ear were arranged together before the two sets were merged. Prints of every twin pair were therefore correctly classified according to the ear that created it. Results for twin pair 1 are illustrated in Figure 9.8. Dendrograms for the other twin pairs are comparable.
___________________________________________________________________________
___________________________________________________________________________ Figure 9.8 Dendrogram indicating the degree of similarity between prints of twin couple 1 determined by a hierarchical cluster analysis using variables derived from the position of seemingly corresponding features. ___________________________________________________________________________
9.4.3 Fully automated comparison of earprints
___________________________________________________________________________ a1 16 4 2 0 0 0 0 0 0 0 0 1 0 0 0 1 0 2 0 2 0 0 1 0 2 0 0 2 0 0 0 1 0 1 1 a2 9 18 1 0 0 0 0 1 0 0 0 0 1 1 0 0 0 1 0 1 0 1 1 0 1 0 0 0 1 0 1 0 0 1 0 a3 11 9 0 3 2 0 0 0 1 2 0 0 0 0 1 1 0 0 0 0 0 0 0 0 1 1 1 0 1 0 0 3 1 1 1 b1 3 0 0 6 37 1 0 2 0 1 0 1 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 2 2 0 1 b2 0 0 3 8 13 1 1 1 1 0 1 0 0 1 0 2 0 2 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 1 1 P a i r 1 b3 0 0 4 32 15 1 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 1 0 0 0 a1 0 0 2 0 0 0 25 27 4 4 4 0 0 1 0 0 1 2 1 0 0 0 1 0 0 0 0 0 0 1 0 1 1 0 0 a2 1 0 0 0 0 0 27 32 5 4 4 0 1 0 0 0 0 2 0 0 1 0 2 0 0 1 0 1 0 0 0 0 0 0 0 a3 0 0 0 0 1 0 34 36 9 8 8 1 0 0 0 1 1 0 0 1 1 0 0 1 1 0 0 0 0 0 0 0 0 0 0 b1 2 0 2 0 0 0 7 7 8 18 17 2 0 0 0 1 0 1 0 0 1 0 0 1 0 1 0 1 2 0 1 1 2 0 0 b2 0 0 0 0 1 0 13 6 4 25 38 0 0 1 0 0 0 0 0 0 0 1 1 0 1 0 0 2 0 0 0 0 1 0 1 P a i r 2 b3 0 0 0 1 0 1 4 8 4 18 36 0 1 1 1 0 0 1 0 0 0 1 0 1 2 0 2 0 0 1 3 2 0 0 1 a1 1 0 1 0 3 0 0 0 0 1 2 0 6 10 2 4 0 0 0 1 0 1 0 0 0 2 0 0 0 1 1 0 0 1 1 a2 0 1 0 0 1 0 0 1 2 2 0 1 8 9 1 4 2 0 2 0 0 0 0 0 1 2 0 0 2 0 0 0 0 1 0 a3 0 0 2 1 1 0 0 0 0 1 0 2 11 14 1 3 0 1 0 0 1 1 0 0 0 0 0 1 1 2 0 1 1 1 2 b1 1 0 2 0 0 0 2 0 0 1 0 0 2 1 1 17 0 1 2 0 2 2 0 0 0 3 3 0 3 0 1 0 0 0 0 b2 1 0 0 0 1 1 0 0 3 0 2 0 3 3 4 16 2 1 1 0 1 4 1 1 1 2 2 0 5 0 1 2 0 1 1 P a i r 3 b3 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 4 11 1 3 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 a1 2 1 1 2 0 1 1 1 0 2 0 2 1 0 0 1 0 0 41 19 1 1 3 1 1 0 0 3 0 0 0 0 3 1 2 a2 0 0 0 1 1 0 2 0 0 1 0 2 1 0 1 0 0 1 50 23 0 3 1 1 1 0 1 0 3 0 1 0 2 1 1 a3 0 1 1 0 0 1 0 1 0 1 1 0 0 0 2 0 0 0 14 28 0 3 1 0 0 1 1 0 2 0 0 1 1 0 2 b1 1 1 1 1 0 0 1 0 0 0 1 0 0 0 0 1 1 0 0 1 0 29 23 0 0 1 2 1 1 0 0 0 1 1 1 b2 0 0 1 0 1 1 0 0 1 1 1 0 0 0 2 0 3 1 1 3 3 28 28 1 0 1 0 2 1 0 0 0 2 0 0 P a i r 4 b3 0 0 1 1 2 0 0 0 0 0 0 2 2 0 1 0 0 0 3 0 2 23 29 1 0 0 0 1 0 0 1 0 1 1 0 a1 1 0 1 1 0 1 0 0 1 1 1 0 2 2 1 0 3 1 2 1 0 0 1 0 14 8 0 1 0 0 0 2 0 1 2 a2 1 0 0 0 0 0 1 1 1 0 1 0 0 0 1 0 1 0 0 2 0 0 1 2 20 22 3 2 1 0 0 0 1 0 1 a3 1 0 0 0 1 1 0 0 0 0 0 0 0 1 0 0 4 0 1 0 1 1 2 0 9 15 0 5 0 1 0 0 0 0 1 b1 0 0 1 1 0 0 0 1 0 0 1 1 0 0 0 2 1 0 1 0 0 1 0 0 0 3 0 13 15 1 0 0 2 0 0 b2 0 0 0 0 1 1 0 0 0 0 0 0 0 2 0 0 1 1 1 0 1 1 0 0 0 2 6 16 28 0 0 1 0 3 0 P a i r 5 b3 0 2 0 1 0 0 1 0 1 3 0 0 1 0 0 0 0 0 1 0 0 0 0 0 1 3 2 12 32 0 1 2 0 0 2 a1 0 1 0 1 0 1 0 0 1 0 1 0 0 0 0 0 1 1 0 0 0 0 1 0 0 0 2 0 0 0 6 8 0 0 1 a2 0 0 1 0 0 0 0 0 0 1 0 0 1 0 1 0 0 1 0 1 0 1 0 0 1 0 1 0 1 0 9 11 2 0 2 a3 0 0 1 0 0 0 0 1 0 0 1 0 1 1 0 0 1 1 0 2 0 1 0 1 0 1 1 2 0 0 7 9 0 1 0 b1 0 0 0 0 0 1 0 1 1 0 1 0 0 1 1 0 0 0 1 0 2 0 1 0 0 1 1 0 0 0 0 1 0 18 17 b2 0 0 0 0 1 0 0 0 0 2 0 0 1 0 0 0 0 1 1 0 0 0 0 2 1 0 0 1 0 0 0 0 1 20 3 P a i r 6 b3 0 0 0 0 0 0 0 0 0 1 0 0 1 0 1 1 1 0 1 2 1 1 0 1 1 1 0 0 0 1 1 4 1 19 5 a1 a2 a3 b1 b2 b3 a1 a2 a3 b1 b2 b3 a1 a2 a3 b1 b2 b3 a1 a2 a3 b1 b2 b3 a1 a2 a3 b1 b2 b3 a1 a2 a3 b1 b2 b3
Pair 1 Pair 2 Pair 3 Pair 4 Pair 5 Pair 6
___________________________________________________________________________ Figure 9.9 Results matrix for 36 earprints showing the number of matching keypoints found for each print comparison. Each of the 6 twin pairs is labelled and for each twin pair there are 6 earprints: the 3 prints from each donor are grouped together. Elements in the matrix showing 5 or more matches are shaded. The diagonal corresponds to a print matched with itself and is therefore left blank.
example, with a (shaded) threshold of 5 keypoint matches or more this is only possible for twin pair 4. Nevertheless, the results are promising given the early stage of development of the method.
Comparisons between prints of non-sibling donors show few keypoint matches. Among this group only 1 from 1080 comparisons showed five keypoint matches or more and no comparison showed more than 5 matches. Conversely, comparisons between prints from the same donor show relatively high numbers of keypoint matches. Among this group 67 from 72 comparisons showed 5 matches or more; the maximum number of matches was 50. Comparisons between prints of one twin member and the other show varying degrees of confusion. Among this group 12 from 108 comparisons showed five matches or more. The twin pair that shows most confusion is pair 2, where 10 out of 18 inter-twin comparisons show 4 or more keypoint matches. However, even for this pair there is a significant difference between the number of matches for inter-twin comparisons (the maximum number of keypoint matches is 13), and the number of matches for intra-donor comparisons (the minimum number of matches is 17).
by suspect X. The denominator is the probability of obtaining a piece of evidence given that the earmark was made by some unknown person. The method presented here represents a step towards the goal of calculating such likelihood ratios. In the equations given by Champod et al. (2001) the nature of the ‘evidence’ is unspecified. However, if we were to take the evidence for two prints being made by the same ear as the number of keypoint matches, we can make statistical estimates of the numerator and denominator in the likelihood ratio. For the method described, the likelihood ratio increases as the number of keypoint matches increases. Since the current sample is small, contains only twins and was collected under controlled conditions, we are only able to put loose and biased estimates on such likelihood ratios. To make more reliable estimates will require large, well-defined samples and this will be addressed in follow-up studies.
9.5 Conclusion
We have presented an analysis of the earprints of 6 pairs of monozygotic twins using different methods with varying degrees of objectivity. After a comparison of prints by eye and by using digital overlays we felt confident that an expert would be able to note diagnostic differences between prints of these twin members, i.e., differences that provide strong evidence for the exclusion of one of the twin members as a possible donor of a print. For two twin pairs (pairs 2 and 4), we feared that this diagnostic clue could occasionally become obscured, depending on the quality of the print (pair 2) or the amount of pressure applied to the surface and/or duration of listening time (pair 4).
For the majority of the studied twin pairs we felt that differences in the positions of imprinted features in prints of different ears – judged from the digital overlays – did not exceed the limits expected for multiple prints of the same ear. The results of the cluster analysis using variables derived from the position of seemingly corresponding features, however, showed that the position of investigated features was more similar in prints of the same ear then between prints of different ears, i.e., different twin members.
unambiguous classification of same-donor prints but it has been used to obtain promising hit-list results. A print from the correct individual is ranked at the top of the hit-hit-list in 36 out of 36 cases. The new method can be used to calculate statistical likelihood ratios for assessing the evidential value of earprint comparisons and should enable objective and repeatable earprint comparisons.
Acknowledgments
We would like to thank the twins that participated in this study for their effort and time. We would further like to thank David Lowe for making the executable code for detecting and describing salient regions available on his web page.
x In our dataset, the positions of common features in earprints of the individual twin members
are different enough to distinguish between the two sets of prints.
x The results of Keypoint Matching, i.e., a similarity measure for any two prints, may be used to
fully automatically rank prints in a database according to their degree of similarity to a query print.
x In our experiment the top-ranked earprint comes from the same individual as the query print
in 36 out of 36 cases.
x Fully automated methods have an important role in allowing objective statistics to be derived
