Image Retrieval from a Forensic Cartridge Case Database - Multimedia Database Retrieval: Technology and Applications - page 160

Table 6.3 Averaged cross-correlation scores for the ten pairs of cartridge cases fired from
ten shotguns

CCF after registration
Without ECA Cartridge cases being correlated CCF of the original With ECA

Code 101 vs. Code 102 0.7972 0.7784 0.9276

Code 201 vs. Code 202 0.4029 0.1632 0.8158

Code 301 vs. Code 302 0.4307 0.2920 0.8540

Code 401 vs. Code 402 0.1567 0.0036 0.8776

Code 501 vs. Code 502 0.3584 0.3025 0.8611

Code 601 vs. Code 602 0.4158 0.5749 0.8724

Code 701 vs. Code 702 0.5140 0.4985 0.8116

Code 801 vs. Code 802 0.0996 0.0139 0.7434

Code 901 vs. Code 902 0.2913 0.0348 0.8123

Code 1001 vs. Code 1002 0.3026 0.3559 0.8874

Average 0 . 3769 0 . 3017 0 . 8463

6.4 Experimental Result

In the experiment, the system performance was measured when the system was
utilizing the registration algorithm, cross-covariance function, and edge mask for
characterization of the ECA. All test fires were taken from twenty test fires (2
cartridges × 10 firearms) that were imaged by 40 × microscopy at a resolution of

480 640 pixels. Each unit had its image measured in the region of the firing-pin

impression and breech-face impression. The ten firearms used in this study are the
same type of shotgun, the same brand of 'Remington', and the same model of 812
with four different serial numbers: C109219, C88922, C123355 and C86487. The
images were indexed according to their classes: images named Code 101 and Code
102 are from class 1, whereas Code 201 and 202 are from class 2, and so on. A
program has been developed, which automates the methods discussed previously,
and produces cross-correlation scores for each pair of cartridge cases. Table 6.3
shows a list of averaged CCF values for all ten pairs of cartridge cases fired from
ten different shotguns. The matching methods that were compared are Method 1:
CCF of original images, as presented in Geradts et al. [ 179 ]; Method 2: CCF of the
original images after registration, as presented in Vorburger et al. [ 168 ]; and Method
3: CCF of the pairs with ECA extraction presented in the current work. Note that
the normalization process discussed in Eqs. ( 6.1 ) and ( 6.2 ) was applied for all three
methods. The CCF value-matching pair is close to one for a perfect match.
For all ten lists, the average CCF obtained by the ECA method had a value of
0.85, where a value of 1.0 would be a perfect score. This matching rate is about 47 %
higher than that obtained with a widely-used system in which the images are directly
correlated without the selection of effective correlation areas and registration [ 179 ],
and about 55 % higher than that of Method 2. This quantity can be used to indicate
the reliability of identifying cartridge cases fired from the same firearm using a

×