J Forensic Sci, September 2013, Vol. 58, No. 5 doi: 10.1111/1556-4029.12224 Available online at: onlinelibrary.wiley.com
TECHNICAL NOTE CRIMINALISTICS Szymon Matuszewski,1 Ph.D.; and Michał Szafałowicz,1 M.S.
A Simple Computer-assisted Quantification of Contrast in a Fingerprint
ABSTRACT: A simple method for quantification of contrast in a fingerprint is proposed. Contrast is defined as the average difference in
intensity of pixels between valleys and ridges in a fingerprint. It is quantified from a scanner-acquired image of the fingerprint using a histogram function of Adobe Photoshop. The method was validated with black inked prints and marks developed with aluminum powder. Moreover, we tested resistance of the method to rater-dependent errors and dependence of the measurements on the resolution of an image and the model of the scanner. For both groups of fingerprints, the method gave coherent and easily interpretable quantitative values for contrast. There were no significant differences between measurements performed by different raters and by the same rater in a test–retest procedure. However, the method was found to be instrument dependent, as measurements were significantly affected by image resolution and the model of the scanner.
KEYWORDS: forensic science, criminalistics, fingerprints, fingerprint quality, fingerprint development, fingerprint image
The contrast between valleys and ridges in a fingerprint is one of the most important parameters of fingerprint quality (1). It may be used to evaluate the performance of new techniques for the development of fingermarks or to make intertechnique comparisons. Techniques for the development of fingermarks give results of differing quality for different kinds of marks, different surfaces, etc. (2). Therefore, contrast may assist in evaluation of development results while validating new techniques or while creating development guidelines for particular kinds of marks, surfaces, etc. Moreover, contrast may have some value for studies on the permanency of fingermarks and age-related changes in fingermarks. Some results suggest that the overall features of marks (e.g., the clarity of marks or the continuity of ridges) show age-related changes and may be of some use for the determination of the age of a mark (3–5). It seems that contrast between valleys and ridges in a fingermark is also related to the age of a mark, so it may be useful for estimating fingermark age. Recently, Humphreys et al. (1) proposed a method for quantifying contrast in a fingerprint. They elaborated a relative contrast index (RCI), which was defined as a log-transformed ratio of the reflective intensity of fingerprint valleys to fingerprint ridges (1). The reflective intensity was measured using a spectrophotometer attached to a microscope, and the method was successfully applied to black inked fingerprints on paper, latent fingermarks on paper developed separately with ninhydrin and physical developer, and fingerprints in blood enhanced with amido black (1). It was also found that different microspectrophotometers produce differing values of RCI (6). In this article, a different approach is proposed and initially validated. It differs from the
1 Marcin Laboratory of Criminalistics, Adam Mickiewicz University, Sw. 90, 61-809 Pozna n, Poland. Received 24 June 2012; and in revised form 6 Sept. 2012; accepted 30 Sept. 2012.
1310
previous method in two main points. First, contrast in a fingerprint is defined as the difference in intensity between valleys and ridges and not as the ratio of intensity. Second, intensity is measured using popular graphics software for personal computers. This approach seems to be simpler and less time and money consuming. Moreover, it gives results which are easier to interpret and more useful for comparisons between different kinds of fingerprints. An Outline of the Method The quantification of contrast as a ratio of intensity in valleys to ridges gives disparate results for fingerprints with dark ridges and light valleys and vice versa (Fig. 1). For this reason, we suggest quantifying contrast as the difference in intensity between valleys and ridges and not as the ratio of intensity. Apart from producing consistent results for dark–light and light– dark fingerprints (Fig. 1), this approach gives results which are more intuitive and easier to interpret. The method uses a popular graphics software (Adobe Photoshop) to get information on the intensity of pixels in an image of a fingerprint. The first step is the acquisition of a digital image for a fingerprint. A flat scanner operating in grayscale mode is used for this purpose. Then, contrast between valleys and ridges is quantified using a histogram function of Adobe Photoshop. The software provides information on the average intensity of pixels within a specified area of an image. So, at least 10 ridges and valleys are selected starting from the core of the fingerprint. Areas of the same size are chosen within these ridges and valleys, and the average intensity of pixels is measured within these areas. Raw differences in intensity between valleys and ridges are quantified, and afterward, the average difference is calculated for the fingerprint. The software represents 256 different intensities (i.e., shades of gray) by numbers, with 0 for ideal black and 255 for ideal white. Consequently, contrast © 2013 American Academy of Forensic Sciences
MATUSZEWSKI AND SZAFAŁOWICZ
FIG. 1––Contrast within a fingerprint quantified as the ratio of intensity in valleys to ridges (R ¼ IIvr ) and as the difference in intensity between valleys and ridges (D ¼ jIv Ir j). A – the mark developed with aluminium powder and preserved using black gelatine lifter; B – the black inked print on white paper. Grey intensity was measured as described in the article. Fingerprints have visually similar contrast; however, comparable numerical values were obtained only by means of measuring the difference in intensity between ridges and valleys.
between valleys and ridges in a fingerprint may take values from 0 (no contrast) to 255 (maximum contrast). In this article, the concept of the method is presented and initially validated. The article also gives results on the consistency of measurements for pictures differing in resolution and pictures acquired with different scanners (intermethod reliability) as well as the consistency of measurements performed by different raters and by the same rater in a test–retest procedure (inter and intrarater reliability).
.
QUANTIFICATION OF CONTRAST IN A FINGERPRINT
humb (Fig. 2). Accordingly, fingerprints of different contrast were obtained. The second kind were lifts repeatedly taken from a single powder-developed fingermark. It was eccrine/sebaceous mark, and it originated from the left thumb of the male. The mark was left on the surface of a glass cup, and it was developed with aluminum powder. Three consecutive lifts were made from the mark using black gelatine lifter. Because the powder was applied only once, the procedure gave three “copies” of the mark which differed in contrast (Fig. 2). The grayscale, 1600 dpi images of fingerprints were acquired with a CanoScan 4200F (Canon Inc., Tokyo, Japan). All builtin scanner corrections were turned off. The average intensity of pixels was measured within a square composed of 121 pixels, as it fitted ridges and valleys in images of 1600 dpi well. The histogram function of Adobe Photoshop CS2 (Adobe Systems Inc. 1990–2005, San Jose, CA) was used for this purpose. We selected fifteen pairs of ridge and valley areas (in every second ridge and following valley starting from the core of a fingerprint). Only those parts of a fingerprint were chosen in which ridges and valleys were clearly distinguishable (e.g., smudged areas were omitted). The average intensity of pixels was quantified for each area, and contrast was calculated for each pair of areas. Afterward, the average contrast for the fingerprint was calculated from these data. We decided to compare three definitions of contrast. The first definition characterizes contrast as the difference in intensity between valleys and ridges (D), where D ¼ jIv Ir j and Iv and Ir are the average intensities of pixels within the valley and the ridge area, respectively. The second definition describes contrast as the ratio of intensity in valleys to ridges (R) where
Materials and Methods Validation of the Concept The method was tested with two kinds of fingerprints. The first kind were prints left on white office paper (A4, 80 g/m2) by a male left thumb, uniformly covered with black ink. Eight prints were deposited one after the other by a single inked t-
1311
R¼
Iv Ir
The third one defines contrast in the manner proposed by Michelson (7) for any patterns with similar portions of dark and light elements (Cm) where
FIG. 2––Contrast in the successively deposited inked prints (upper row) and consecutive lifts from the single aluminium-powdered mark (lower row). In the case of inked prints, the 2nd and the 6th depositions were omitted. Contrast was quantified as the difference in intensity of pixels between valleys and ridges (D), as the ratio of intensity in valleys to ridges (R) and as the Michelson contrast (Cm). Fingerprints were arranged according to the value of D.
1312
JOURNAL OF FORENSIC SCIENCES
Cm ¼
jIv Ir j Iv þ Ir
In all experiments, subjects were pairs of ridge and valley areas. We accepted a 5% level of significance. Calculations were made using Statistica 9.1 (StatSoft Inc. 1984–2010, Tulsa, OK).
Intermethod Reliability
Results and Discussion
To test whether the resolution of an image affects measurements of contrast, we scanned the powder-developed fingermark using resolutions of 400, 800, 1200, and 1600 dpi. Contrast was defined as the difference in intensity between valleys and ridges and was measured as described in the previous part of this section. The same ridge and valley areas were chosen in each image. Measurements of contrast for each pair of areas were subsequently used as a sample for a given resolution and were included in statistical analyses. Because contrast was measured for the same areas in each image, the significance of differences between images was tested with ANOVA for repeated-measures designs and image resolution as a within-subject factor. To test differences in measurements on images from different scanners, we scanned the powder-developed fingermark using a CanoScan 4200F (Canon Inc.), a CanoScan Lide110 (Canon Inc.) and a Microtek 3830 (Microtek International Inc., Hsinchu, Taiwan). Scanners were working in grayscale mode with all corrections turned off. A resolution of 1200 dpi was used. Contrast was defined and measured according to the same procedure as in the resolution experiment. ANOVA for repeated-measures designs was used with the model of the scanner as a within-subject factor.
Validation of the Concept Numerical values were coherent and easily interpretable only in the case of contrast quantified as the difference in intensity between valleys and ridges (Fig. 2, Table 1). Arrangement of fingerprints according to these values gave a sequence which consistently reflected a decrease in the quality of fingerprints (Fig. 2, Table 1). The other definitions resulted in similar values for fingerprints with clearly different contrast, as for example in the case of aluminum 1 and 2 and the “ratio” definition or aluminum 3 and Inked 1 and Michelson contrast (Fig. 2, Table 1). These results support the superiority of the “difference” definition over other definitions. They also demonstrate that the concept of the method is valid. Average contrast quantified as the difference between ridges and valleys ranged from 5 to 115 intensity points (Table 1). For high-quality fingerprints, it was above 90, although at values above 20 points, ridges were still easily recognizable (Fig. 2). In all fingerprints, there was substantial variation in contrast (Table 1). Accordingly, while measuring contrast, one should take into account an appropriate number of ridges and valleys, that is, no less than 10 pairs from the average area of a fingerprint. The level of variation increased with the decrease in the average contrast (Table 1). This indicates that decrease in quality is not uniform for the fingerprint.
Inter- and Intrarater Reliability The consistency of measurements between and within individuals was evaluated for two raters and three sets of measurements. A CanoScan 4200F (Canon Inc.) was used with a resolution of 1600 dpi. Raters were males with experience of work with fingerprints. However, they were not fingerprint identification experts. Raters quantified contrast three times on the same image, each time for 15 prespecified pairs of ridges and valleys. Consequently, six samples of measurements were included in ANOVA for repeated-measures designs with replication of measurements as a within-subject factor.
Intermethod Reliability Not surprisingly, image resolution significantly affected measurements of contrast (rmANOVA, F3,42 = 8.4, p = 0.00017), with higher resolutions resulting in higher contrast (Fig. 3). However, differences in contrast were rather small (for example, the difference between images at 400 and 1600 dpi was about 8 intensity points). It was difficult to demarcate ridge and valley areas in images of low resolution (400 and 800 dpi). For that
TABLE 1––Contrast quantified as the difference in intensity between valleys and ridges (D), as the ratio of intensity in valleys to ridges (R) and as the Michelson contrast (Cm). Variation in Contrast within a Fingerprint [the Coefficient of Variation; %]
The Average Contrast for a Fingerprint
Inked 1 2 3 4 5 6 7 8 Aluminum 1 2 3
D
R
Cm
D
R
Cm
114.6 99.9 65.6 36.3 20.7 17.6 13.0 8.7
1.96 1.76 1.38 1.18 1.09 1.08 1.06 1.04
0.32 0.27 0.16 0.08 0.04 0.04 0.03 0.02
10.1 12.6 13.3 23.3 27.9 22.9 22.0 44.0
9.3 9.2 5.1 4.1 2.5 1.9 1.2 1.6
13.2 15.5 15.0 24.5 28.5 24.6 22.2 44.1
90.9 30.3 5.4
0.14 0.16 0.54
0.75 0.72 0.30
7.4 20.7 53.2
14.8 26.9 28.8
4.2 8.7 43.0
Inked 1–8 – prints deposited one after the other with the single inked finger. Aluminum 1–3 – lifts repeatedly taken from the single powder-developed mark.
MATUSZEWSKI AND SZAFAŁOWICZ
FIG. 3––Measurements of contrast for images differing in resolution. Contrast was defined as the difference in intensity of pixels between valleys and ridges. Means (points) and their 0.95 confidence intervals (vertical bars) are presented.
.
QUANTIFICATION OF CONTRAST IN A FINGERPRINT
1313
FIG. 5––Measurements of contrast replicated three times by two raters (SM1-3, MS1-3). Contrast was defined as the difference in intensity of pixels between valleys and ridges. Means (points) and their 0.95 confidence intervals (vertical bars) are presented.
Concluding, in this article, we have proposed and validated a novel method for quantification of contrast in a fingerprint. Within this technique, contrast is defined as the average difference in intensity of pixels between valleys and ridges in a fingerprint, and it is quantified using a scanner-acquired image and a histogram function of Adobe Photoshop. The method gave coherent and easily interpretable results for black inked prints and aluminum-powdered marks. It was also found that the method is instrument dependent and resistant to rater-dependent errors. Acknowledgments We thank the anonymous reviewers for helpful comments and suggestions on earlier versions of the manuscript. References FIG. 4––Measurements of contrast for images acquired using different scanners. Contrast was defined as the difference in intensity of pixels between valleys and ridges. Means (points) and their 0.95 confidence intervals (vertical bars) are presented.
reason, it is recommended that contrast is quantified in images of fingerprints scanned with a resolution of at least 1200 dpi. Measurements of contrast from images acquired with different scanners were significantly different (rmANOVA, F2,28 = 61.9, p < 0.000001; Fig. 4). Similar results were demonstrated by Vanderwee et al. (6) for the relative contrast index, as measured with a microspectrophotometer. Therefore, it is evident that measurement of contrast in a fingerprint is instrument dependent. Inter- and Intrarater Reliability There were no significant differences between measurements performed by different raters on different occasions (rmANOVA, F5,70 = 0.56, p = 0.73; Fig. 5). This consistency indicates that the method is resistant to rater-dependent errors.
1. Humphreys JD, Porter G, Bell M. The quantification of fingerprint quality using a relative contrast index. Forensic Sci Int 2008;178:46–53. 2. Lee HC, Gaensslen R. Methods of latent fingerprint development. In: Lee HC, Gaensslen R, editors. Advances in fingerprint technology. Boca Raton, FL: CRC Press, 2001;105–75. 3. Barnett PD, Berger RA. The effects of temperature and humidity on the permanency of latent fingerprints. J Forensic Sci Soc 1977;16:249–54. 4. Baniuk K. Determination of age of fingerprints. Forensic Sci Int 1990;46:133–7. 5. Champod C, Lennard C, Margot P, Stoilovic M. Fingerprints and other ridge skin impressions. Boca raton, FL: CRC Press, 2004. 6. Vanderwee J, Porter G, Renshaw A, Bell M. The investigation of a relative contrast index model for fingerprint quantification. Forensic Sci Int 2011;204:74–9. 7. Michelson A. Studies in optics. Chicago, IL: University of Chicago Press, 1927. Additional information and reprint requests: Szymon Matuszewski, Ph.D. Laboratory of Criminalistics Adam Mickiewicz University Marcin 90 Sw. 61-809 Poznan Poland E-mail:
[email protected]
