Design of an optical sensor to improve the detectability of pores in fingerprints

After the September 11th 2001 incident, the application of biometrics is a fast growing business. Essentially, a biometric system is a pattern recognition system that operates by acquiring biometric data from an individual, extracting a feature set from the acquired data, and comparing this feature set against the template set in the database. The technology relies on the automatic assessment of a unique body feature, such as a hand, face, ear, voice, odour (smell), gait, iris, DNA or fingerprint. Depending on the application context, a biometric system may operate either in verification or in identification mode. There are a number of biometric based identification and verification systems available on the market, mainly for military and forensic applications, with the main emphasis on identification. Financial institutions like banks, however, are seeking for biometric alternatives for verification purposes, in order to replace the commonly used PIN (Personal Identification Number). Only a few biometric technologies (iris, retina, DNA and fingerprints) may fulfil this requirement of the banks. Combined with a patent submitted by Dr. L.J. van Ruyven under patent number WO 93/18486, the TNO Research tests on a specifically, by Siemens, developed elastomer for this purpose, this project came to life. A comprehensive study has been performed on presently available biometric identification and verification devices, evaluating the pro’s and cons. Collectability, acceptability and fraud sensitivity (resistance) pushed this research to the application of fingerprint verification only. Basically, there are two rules on which the science of fingerprint verification and identification is based on: 1. The fingerprints are "permanent" in that they are formed prior to birth, and remain the same throughout lifetime, until sometime after death when decomposition sets in. 2. The fingerprints are "unique"; no two fingerprints, or friction ridge areas, made by different fingers or areas, are the same or are identical in their ridge characteristic arrangement. Fingerprints can be classified in three levels; level 1 classification by fingerprint patters, level 2 classification by specific characteristics, like minutiae points such as bifurcations, ridge ends and dots and finally level 3 classification by dimensional attributes of a ridge, such as a ridge path, width, shape, pores, edge contour, incipient ridges, breaks, creases, scars, and other permanent details. The latter one is called high-resolution features. The specific level 2 features were detected by Sir Francis Galton and are therefore called Galton features. There are 13 different Galton features classified. It should be noted that the uniqueness of the fingerprint (set of papillary ridge lines) does not automatically imply the uniqueness of a set of features of a fingerprint. As, in general, only a portion of the entire fingerprint is investigated, the uniqueness of a set of features has to be proven. A portion of a fingerprint is taken and divided into cells with a dimension of 1 mm x 1 mm, whereby the frequency of occurrence of the 13 possible Galton features is tested. In our specific fingerprint set of features, we calculated the probability of a configuration of 16 minutiae in an observed area of 49 mm2, of 2.492 x 10-23. This outcome confirms the uniqueness of fingerprint feature configurations, necessary for the next step of the research. A patent was used, based on the assumption that persons could be identified by the geometrical property of distances between the papillary ridges. The combination with the elastomer foil should facilitate the performance of ridge tracking. The result was only partly acceptable for level 2 verification, as the foil on the applied press-plate is so stiff that only lines, bifurcations and endpoints of the fingerprint can be detected. It became clear that pores could not be detected with the applied stiff press-plate. The next step was to perform additional simulations, Finite Element Analysis, in order to check whether a press plate could be used at all for pores detection. A thinner and less stiff foil has been considered, applying a gel instead of an elastomer. The tests showed, however, that a weaker press-plate would result in a very vulnerable, almost unusable system. An even greater problem occurred by assuming that the ridge structure would behave like congruent images under all environmental conditions. The impact of Humidity and Temperature were tested separately to verify this assumption, but most of all to find a mathematical relation between the obtained images. It was concluded that the obtained images do change under different environmental conditions and that mathematical compensation will not solve this problem under all possible conditions. Therefore, the assumption of compensating congruent images is rejected. The final test, the impact of pressure, based on the mechanical properties of the skin, showed significant changes in ridge structures under different pressure and turned out to be the main reason for rejecting the usage of finger line tracking based on the patent. As the initial project description was abandoned, a different prototype design was developed and built, using a scanning technique of an orthogonal grid. The choice for this technique was mainly based on new results from literature research. With scanning in a grid all the necessary information of the fingerprint could be retrieved. The constructed prototype consists of an optical system with the press-plate, drives, data-acquisition equipment, reconstruction algorithm and the necessary interfaces. An altitude map chart, containing sufficient papillary ridge information was obtained for both slope directions (x- & y- direction). Initially, it was not possible to synchronise these two independent measurement data, a minor shift occurred. The obtained shift was caused by the not constant rotating speed of the rotating mirror. By applying two trigger signals, better results were obtained. By building this prototype, it was proven that images could be made and that the reconstruction algorithm for transforming the obtained slope information into an altitude chart is possible. Level 2 features are detectable. Nevertheless, this does not fulfil the requirement to distinguish pores. All the above-performed steps made clear that level 2 classifications could not suffice. An additional feature is required and pores seemed to be the most suitable level 3 add on. The characteristics of pores and its spatial distribution were investigated, showing the uniqueness of intra-ridge pores configurations. Therefore, adding pores to the standard level 2 classification techniques could result in the fulfilment of the bank requirements. The initial system architecture was adapted accordingly. By adding pores the required surface could be diminished, but the scanning resolution should be increased. In general, at level 2 verification, a scanning resolution (Rlevel.1 = Rlevel.2) of 20 points or pixels per mm is sufficient. As pores should be detected, the sampling period of half the size of the smallest pore, 60 µm, is applied. This results in a minimum resolution of the sensor of approximately 33 points per mm. A new prototype was developed using no moving parts and consisting of the following components: • Ring green LED illumination; using strike light instead of direct illumination. This application resulted in a substantial higher contrast of the image. • Telecentric lens; a telecentric objective certifies the same magnification when small distance variations in the axis direction occur. Therefore the position of the finger may vary slightly in that direction. • CCD camera; the spatial resolution, the sampling rate, is approximately 120 ppmm in the horizontal direction and 116 ppmm in the vertical direction. This is 3,75 times greater than the required resolution of the sensor. • Ring holder for the positioning of the finger; to assure that the position of the finger is almost similar at all circumstances, the choice is made for a ring, a kind of aperture, with a inner diameter of approximately 7 mm. Furthermore a ring has advantages over a standard glass device. As the reasons for these choices are comprehensively described in chapter 3 of this thesis, some unique features should be highlighted. The present available sensors mostly have direct light. By applying ring LEDs (Light Emitting Diodes) strike light is obtained, improving the contrast ratio. The choice of a different green or blue wavelength instead of standard white light is also a unique additional feature. It is related to the penetration level of the light into the skin, in order to obtain the maximum reflection. To test the performance of all properties of the applied components of the set up, the optical design software program Zemax has been applied. These simulation tests have shown that this sensor set up fulfils the initial bank requirements. The chosen camera and the applied telecentric lens system meet the required MTF to obtain a sufficient contrast to distinguish the minimum size of a pore for level 3 classification. The field curvature of the sensor, as function of the distance to the optical axis is low, maximum 27 µm, and can be neglected. The depth of field shows that the image of the finger is still sharp with a shift of the finger of maximum 10 mm and an out of focus direction of maximum 600 µm. Vignetting, the loss of light, when the beam that enters the objective at an angle with the optical axis may miss a part of the second lens, or the chip of the camera, is less then 3-4 % and therefore negligible. Spherical aberration, coma and astigmatism are also negligible. It is concluded that the obtained images show distinctive pores, the main objective of the research. The final step in the design of an automatic fingerprint recognition system is the performance assessment of the system. The objective was to maximise the acquisition process, which mainly determines the performance of the complete system. A homemade software algorithm was added to combine two different, generally used, techniques based on grey value algorithms for level 2, ridges and minutiae, extraction and adapted thinning for level 3, pores extraction. A specific test image was used with distinct pores. This test image has almost negligible noise. The results of this extraction algorithm were used to match and compare with the fingerprint features statistical analysis and system performance estimates, as described by Roddy and Stosz [RS97]. The performance of a system is determined and judged by the feature uniqueness or variations of matching parameters, in other words by the False Acceptance Rate (FAR) and the False Rejection Rate (FRR). The FAR is directly related to the feature uniqueness of a configuration, the feature area, the number of features and the density of features. The FRR focuses on the inherent feature reliability, pores visibility, and the efficiency of the feature detection algorithm. The feature uniqueness has been proven for the sensor for different pores configurations, supporting the assumption of Ashbaugh and Locard [Ash95, Loc12], that 20 pores are sufficient to identify or verify a person. For the feature inherent reliability (Ri) and the algorithm detection reliability (Rd), separate methods have been applied. Ri has been determined empirically, as mathematical methods will generate algorithm errors. Algorithm detect reliabilities are determined by the missed detects and the false detects. All these determined reliabilities are combined to achieve the performance characteristics of our sensor, in other words, the FAR and FRR is determined, as function of the match score. Projecting these results on the required specifications of the bank, a FAR of 0,01% and a FRR of 0,005%, the outcome is above expectation. The observed performance of the prototype sensor meets the performance specifications of the banks by far.