Biometrics Technology: Finger Prints

Transcription

1 References: Biometrics Technology: Finger Prints [FP1] L. Hong, Y. Wan and A.K. Jain, "Fingerprint Image Enhancement: Algorithms and Performance Evaluation", IEEE Trans. on PAMI, Vol. 20, No. 8, pp , August [FP2] A.K. Jain, L. Hong and R. Bolle, "On-line Fingerprint Verification", IEEE Trans. on PAMI, Vol. 19, No. 4, pp , [FP3] A. K. Jain, S. Prabhakar, L. Hong and S. Pankanti, "Filterbank-based Fingerprint Matching", IEEE Transactions on Image Processing, Vol. 9, No.5, pp , May [FP4] A.K. Jain, S. Pankanti, A Touch of Money, IEEE Spectrum, July [FP5] Davide Maltoni, Dario Maio, Anil K. Jain and Salil Prabhakar, Handbook of Fingerprint Recognition, Springer, (ISBN ). [FP6] Lawrence O Gorman, FingerPrint Verification, Chapter 2, Bometrics: Personal Identification in Networked Society, Anil Jain, Ruud Bolle, Sharath Pankanti, Ed. Kluwer Academic Publishers, w05-fingerprints Biometrics - Summer

2 Biometrics Technology: Finger Prints Outline: History (ref.[fp6]) Acquisition Devices (Ref.[FP6]) Raw data Features, representation (Ref.[FP1, FP2, FP3]) Matching (Ref.[FP1, FP2, FP3]) Application (Ref.[FP4]) w05-fingerprints Biometrics - Summer

3 Why Fingerprints? Fingerprint-based identification has a long history. Different fingers have different ridge characteristics (minute details). Minute details are permanent. Fingerprint identification is acceptable in a court of law. Fingerprint on Palestinian lamp (400 A.D.) Bewick s trademark w05-fingerprints Biometrics - Summer

4 History The oldest form of biometrics. There is archaeological evidence that ancient Assyrians and Chinese had used fingerprints as a form of identification. Many impressions of fingers are found on ancient pottery, as on Roman tiles Clay pottery from these times contain fingerprint impression to signify the potter. Clay seals marked by thumbprints are found in Chinese documents. In the mid 1800 s, scientific studies established two critical characteristics of fingerprints: 1. No two fingerprints from different fingers have been found to have the same ridge pattern; and 2. Fingerprint ridge patterns do not change throughout life. w05-fingerprints Biometrics - Summer

5 History Use of fingerprints for criminal identification: 1896 Argintina, South America 1901 Scotland Yard, UK 1900 s to todate other countries Computer processing of fingerprints began in the early In 1980, personal computers and optical scanners enabled the capturing of fingerprint practical in a non-criminal setting. w05-fingerprints Biometrics - Summer

6 History Grew (1684) published the first scientific paper on ridges, furrows and pore structures Mayer (1788) gave a detailed description of the anatomical formation of fingerprints Bewick (1809) began to use his fingerprint as his trademark Purkinje (1823) classified fingerprints into nine categories based on ridge configurations Herschel (1858) started using fingerprints instead of signatures on legal contracts in Bengal Henry Fauld (1880) suggested "scientific identification of criminals" using fingerprints Galton(1888) introduced the minutiae features for fingerprint matching. w05-fingerprints Biometrics - Summer

7 History 1899, Edward Henry established the well-known Henry system of fingerprint classification. The biological principles of fingerprints are summarized: (p.23-24, of [FP5]) 1. Individual epidermal ridges and furrows have different characteristics for different fingerprints established the foundation of fingerprint recognition 2. The configuration types are individually variable, but they vary within limits that allow for a systematic classification established the foundation of fingerprint classification 3. The configurations and minute details fo individual ridges and furrows are permanent and unchanging. FBI (1924) set up a fingerprint identification division with a database of 810,000 fingerprints w05-fingerprints Biometrics - Summer

8 Formation of Fingerprints ref.[fp5] They are fully formed at about 7 months of fetus development. Finger ridge configurations do not change throughout the life of an individual except due to accidents. Appearance and fingerprints are a part of an individual s phenotype (genetic and environment) The general characteristics of the fingerprint emerge as the skin on the fingertip begins to differentiate. The differentiation process is triggered by the growth in size of the volar pads on the palms, fingers, soles and toes. However, the flow of the amniotic fluids around the fetus and its position in the uterus change during the differentiation process. w05-fingerprints Biometrics - Summer

9 Formation of Fingerprints ref.[fp5] Thus the cells on the fingertips grow in a microenvironment that is slightly different from hand to hand and finger to finger.... A small difference in micro-environment is amplified by the differentiation process of the cells. There are so many variations during the formation of fingerprints that it would be virtually impossible for two fingerprints to be exactly the same. But, because the fingerprints are differentiated from the same genes, they are not totally random patterns either. By definition, identical twins cannot be differentiated based on DNA. Typically, most of the physical characteristics such as body type, voice and face are very similar for identical twins. w05-fingerprints Biometrics - Summer

10 Formation of Fingerprints ref.[fp5] Although minute details in the fingerprints of identical twins are different, a number of studies have shown significant correlation in the fingerprint class of identical twin fingers. In dermatoglyphics studies, the maximum generic difference between fingerprints has been found among individuals of different races. Unrelated persons of the same race have very little generic similarity in their fingerprints, Parent and child have some generic similarity as they share half the genes, Siblings have more similarity, and the maximum generic similarity is observed in identical twins. w05-fingerprints Biometrics - Summer

11 The main parameters: Acquisition Devices Resolution the no. of pixels per inch (dpi). FBI requires 500 dpi. Area the larger the area, more patterns can be captured. Typically, an area of 1x1 square inch is sufficient. Number of pixels simply the product of resolution and area, e.g. 400x300 is not uncommon. Dynamic range (or depth) the intensity of each pixel, 8 bits (256 gray levels) is common. (no need of color) Geometric accuracy depending on the relative positions between the finger and the sensor, how much distortion is introduced and/or recovered. Image quality depends very much on the intrinsic finger quality of the user. For example, the patterns may be very light, fingers may be too wet, or too dry. w05-fingerprints Biometrics - Summer

12 Fingerprint sensors w05-fingerprints Biometrics - Summer

13 Finger Print Patterns w05-fingerprints Biometrics - Summer

14 Inking - (Off-line) Acquisition Device traditional mode of criminal fingerprint capture before advances in electronic H/W inconvenient for automatic verification due to the inconveniences involved, i.e. ink and subsequent digitization of the imprint. Connotation of inking fingerprints implies criminal activities not too welcome!! Quality is not good usually. w05-fingerprints Biometrics - Summer

15 Optical sensors Life Scan Devices Ref.[FP5] Frustrated Total Internal Reflection (FTIR) (Fig. 2.6) FTIR with a sheet prism (Fig. 2.7) Optical Fibers (Fig. 2.8) Electro-optical (Fig. 2.9) Direct reading uses a high-quality camera. No contact. But obtaining well-focused and good quality image is very difficult. Solid-state sensors (Silicon Sensors) Capacitive (Fig.2.10) Thermal Electric field Piezoeletric Ultrasound sensors (Fig. 2.11) Sweep Types (section 2.5) Images are constructed from the slices captured by the sensors. Initially as necessary in the thermal device. w05-fingerprints Biometrics - Summer

16 Optical Sensor Frustrated Total Internal Reflection Features Finger touches the top of the glass prism, i.e. ridges in contact with the glass Left side of the prism is illuminated The light entering the prism is reflected at the valleys The light rays exit from the right side of the prism and are captured via CCD or CMOS image sensor Fig.2.6 FTIR-based fingerprint sensing [FP5] Limitation: Trapezoidal distortion w05-fingerprints Biometrics - Summer

17 Optical Sensor FTIR with a sheet prism Features An array of microprisms is mounted upon an elastic surface. When a fingerprint is applied to the surface, the different ridge and valley pressures alter the planar surfaces of the micro-prisms. This image is captured optically via the reflected light (or the absence of it) from the micro-prisms. Fig.2.7 FTIR-based with sheet prism [FP5] Limitation: Image quality is lower than glass prism w05-fingerprints Biometrics - Summer

18 Optical Sensor Optical Fibers Fig.2.8 Based on Optical Fibers [FP5] Limitation: Higher cost Features Substituting prisms with fiber-optic platen, significant reduction in packaging can be achieved A bundle of optical fibers is aimed perpendicularly to the fingerprint surface. These illuminate the fingerprint and detect reflection from it to construct the image. The CCD/CMOS is in direct contact with the platen w05-fingerprints Biometrics - Summer

19 Optical Sensor Electro-optical Fig.2.9 Electro-optical sensor [FP5] Limitation: Quality of image is lower than that from FITR Features consists of 2 layers: Polymer layer when polarized with proper voltage, emits light that depends on the potential applied on one side. As ridges touch polymer while valleys do not, the potential is different, the amount of light emitted also varies a luminous representation of the fingerprint pattern to be generated 2 nd layer - photodiode array embedded in glass captures the pattern as image. w05-fingerprints Biometrics - Summer

20 Life Scan Devices Ref.[FP5] Solid-state sensors (Silicon Sensors) Attempt to reduce the size and cost of optically based sensor. However, the cost of silicon sensors is not less because a smaller area of sensing is not acceptable. Consist of arrays of pixels of sensors. The user touches directly the surface of the silicon surface Neither optical components or external CCD/CMOS image sensors are needed. Electrical signals are generated to capture the ridges and valleys of the finger. Four main types to convert physical information to electrical signals: Capacitive (Fig.2.10) Thermal Electric field Piezoeletric w05-fingerprints Biometrics - Summer

21 Solid-state sensors - Capacitive Fig.2.10 Capacitive Sensing [FP5] Limitation: The need to clean the surface frequently to prevent grease and dirt on the surface. Features A 2-D array of microcapacitor plates embedded in a chip. The other plate is the finger skin itself. Small electrical charges when the finger is in contact with the plate. Critical component is the surface coating as thin as possible but able to be resistant to physical abrasion. need to reduce the sensitivity to electrostatic discharges, chemical corrosion, etc. w05-fingerprints Biometrics - Summer

22 Life Scan Devices Ref.[FP5] Solid-state sensors (Silicon Sensors) Thermal: Made of pyro-electro material that generates current based on the temperature differentials. Ridges and valleys on fingerprints produce different temperatures. Advantage: not sensitive to ESD, can accept a thick protective coating. Electric field: Sensor consists of a drive ring and a matrix of active antennas. The finger must be in contact with the sensor such that the analogue response of each element in the sensor matrix is amplified, integrated and digitized. Piezoelectric: The surface is made of non-conducting dielectric material which on encountering pressure will generate electric current. Ridges and valleys generate different amounts of current due to pressure applied. Limitations: materials not sensitive to produce sufficient current to detect differences use of micro-switches; coating is still a problem also produces a binary image; w05-fingerprints Biometrics - Summer

23 Life Scan Devices - Ultrasound Sensor Fig Basic principle of ultrasound sensing [FP5] Limitation: scanner is large in size; expensive; takes a few seconds to capture an image. Features Based on acoustic signals (echography): Acoustic signals are sent towards the finger and the echo signals are captured. The echo signal is used to compute the image of the ridges and valleys. Two components: transmitter and receiver. Advantages: resilient to dirt and grease on the fingers; works even when user wears thin gloves w05-fingerprints Biometrics - Summer

24 Table 2.1 [FP5] w05-fingerprints Biometrics - Summer

25 Raw Data - Fingerprint Images A rolled inked fingerprint Digital Biometrics sensor (508x480) Fidelica sensor (256x256) Veridicom sensor (300x300) w05-fingerprints Biometrics - Summer

26 Raw Data - Fingerprint Images Arch (A) Tented Arch (T) Right Loop (R) Left Loop (L) Double Loop (W) Whorl (W) w05-fingerprints Biometrics - Summer

27 Raw Data - Fingerprint Images Fingerprints from two different fingers Fingerprints from the same finger w05-fingerprints Biometrics - Summer

28 Finger Print Image Enhancement ([FP1]) In general, due to skin conditions (e.g. dry, wet, bruise, etc.), sensor noise, incorrect finger pressure, and inherent low quality fingers, many fingerprints acquired are of low quality. Enhancement is necessary [FP1] Pre-processing the images to binary form, directional, etc. are also necessary. Enhanced w05-fingerprints Biometrics - Summer

29 FingerPrint - Features The most evident structural characteristics of a fingerprint is a pattern of interleaved Ridges lines that flow in various patterns (dark) Vary in width from 100 μm (very thin) to 300 μm (very thick) Valleys spaces between the ridges (white) Ridges and Valleys usually run in parallel, with a cycle period of about 500 μm They bifurcate or terminate Ref.[FP4], Chapter 3 w05-fingerprints Biometrics - Summer

30 FingerPrints Feature Extraction At the global level, ridge lines assumes distinctive shapes within regions. These regions (singularities) are typically classified into 2 topologies: core and delta core delta Whorl (W) Tented Arch (T) w05-fingerprints Biometrics - Summer

31 FingerPrint - Features delta delta Arch (A) core Tented Arch (T) core Right Loop (R) delta? core Left Loop (L) core Double Loop (W) Whorl (W) w05-fingerprints Biometrics - Summer

32 FingerPrint - Features At the local level behavior of the ridges provides more details. Different ways that the ridges becomes discontinuous is referred to as minutia (small details). A ridge can suddenly come to an end, i.e. terminate, or A ridge can divide into two ridges, i.e. bifurcate. (Figure 1 of [FP1]) Ridge Bifurcation Ridge Ending w05-fingerprints Biometrics - Summer

33 w05-fingerprints Biometrics - Summer

34 Minutiae Extraction 1. Input Image 2. Orientation Estimation 3. Ridge Filter 5. Postprocessing 4. Ridge Thinning 6. Minutiae Extraction w05-fingerprints Biometrics - Summer

35 Estimation of Orientation Field Ref. FP[1] Orientation is estimated in within windows, e.g. 16x16 Refined using local orientation consistency Directional image Before refinement After refinement w05-fingerprints Biometrics - Summer

36 Ridge Extraction Input image Extracted ridges w05-fingerprints Biometrics - Summer

37 Post-Processing - Ridge w05-fingerprints Biometrics - Summer

38 Minutiae Detection [FP2] Input live-scan image Extracted minutiae: Position (x,y) Direction (θ) w05-fingerprints Biometrics - Summer

39 Minutiae Verification and Classification Minutia detection without pruning Results of minutia verification (rejected=yellow) Minutia classification (bifurcation=green; ending=red) w05-fingerprints Biometrics - Summer

40 Fingerprint Matching Ref.[chapter 4, FP5] This is an extremely difficult problem, due to the large variability in the different images of the same finger, i.e. large intra-class variation. Factors causing the difficulty include Displacement Rotation Partial overlap Non-linear distortion Pressure and skin condition Noise Feature extraction errors w05-fingerprints Biometrics - Summer

41 Fingerprint Matching Ref.[chapter 4, FP5] In general, there are three categories of matching methods: 1. Correlation-based two images are superimposed. The correlation between corresponding pixel taken as the matching score. 2. Minutiae-based [FP2] most popular and widely used technique. Alignment between the template and input minutiae set indicates the matching score 3. Ridge feature-based minutiae features are difficult to extract in poor quality images. This approach compares features extracted from the ridge patterns, e.g. delta, core, furrow, etc. or texture features [FP3] w05-fingerprints Biometrics - Summer

42 Correlation-based Fingerprint Matching Ref. [Section 4.2, FP5] Let T and I the template and input fingerprint image respectively, then the intuitive measure of their diversity (distance) is the sum of the squared differences (SSD) between the intensities of the corresponding pixels, i.e. SSD ( T, I ) = T I 2 = ( T I ) T ( T I ) = T 2 + I 2 2T T I. The terms T 2 and I 2 are constants, the diversity between the two images is minimized when the cross-correlation (CC) between T and I is maximized, CC T ( T, I ) = T I. Note that the quantity (-2) CC(T,I) appears as the third term in SSD. Thus, CC is a similarity measure!! w05-fingerprints Biometrics - Summer

43 Correlation-based Fingerprint Matching Ref. [Section 4.2, FP5] Unfortunately, due to displacement and rotation, their similarity cannot be simply the CC!! Rotational and displacement effect must be incorporated in the similarity measure. Let θ be the rotation and (Δ x, Δ y ) be the displacement of the input image I repsectively, then the similarity between the two images can be measured as S ( T, I ) = max Δ x, Δ y, θ CC ( T, I ( Δ x, Δ y, θ ) ). Due to the non-linear distortion, skin condition and finger pressure, it s rarely good result can be obtained from the direct application of the equation. In addition, direct application of the equation is computationally costly. There are much research activities in this type of matching. w05-fingerprints Biometrics - Summer

44 Correlation-based Fingerprint Matching Ref. [Section 4.2, FP5] Fig. 4.3 (p.139) 2 Impressions of same finger, and the best alignment (max. correlation) a) Very similar, correlates well. b) High distortion, c) Bad skin condition w05-fingerprints Biometrics - Summer

45 w05-fingerprints Biometrics - Summer Minutiae-based Fingerprint Matching Ref.[FP2, chapter 4, FP5] Most well-known and widely used method todate. In this approach, the fingerprint representation is a feature vector (of variable length) of minutiae. Most minutia is represented by a triplet, m={x,y,θ } where (x,y) gives the location of the minutia, and the angle. Thus, template, T and input, I of two fingerprints are represented by. 1,..., }, ', ', ' { ' }, ',..., ', ' {. 1,..., },,, { },,...,, { n j y x m m m m I m i y x m m m m T j j j j n i i i i m = = = = = = θ θ Two minutiae considered to be matched, if the spatial distance (sd) and the directional difference (dd) are smaller than some threshold values, i.e.. ) ',360 ' min( ), ' (, ) ' ( ) ' ( ), ' ( θ θ θ θ θ = + = i j i j i j i j i j i j m m dd r y y x x m m sd

46 Minutiae-based Fingerprint Matching Ref.[FP2, chapter 4, FP5] Based upon this idea, many researchers have developed different algorithms to optimize the matching. Pictorially, The o are the T minutiae, and The x are the I minutiae. w05-fingerprints Biometrics - Summer

47 Another Minutiae Matching Algorithm Ref.[FP2] Generate Alignment Hypothesis Find Similarity Using Elastic String Matching Matching score w05-fingerprints Biometrics - Summer

48 Minutiae Matching Result w05-fingerprints Biometrics - Summer

49 Ridge Feature-based Matching Ref.[FP3] Reference point(x), the region of interest and 80 sectors superimposed on a fingerprint. w05-fingerprints Biometrics - Summer

50 w05-fingerprints Biometrics - Summer

51 Decomposition using Gabor Filters 45 o orientation filter 90 o orientation filter 45 o filtered image 90 o filtered image w05-fingerprints Biometrics - Summer

52 640-dimensional FingerCode Finger 1, Impression 1 Finger 1, Impression 2 Finger 2, Impression 1 Finger 2, Impression 2 FingerCode FingerCode FingerCode FingerCode w05-fingerprints Biometrics - Summer

53 New Approach - Fingerprint as a range image (a) Input Image (b) Median filtered Image (c) Segmented Image (d) Range image w05-fingerprints Biometrics - Summer

54 Another matching algorithm - Mosaicking Algorithm First image Mosaicked image Second image Final alignment of images Augmented minutiae set w05-fingerprints Biometrics - Summer

55 Fingerprint Alignment T1-1 T1 T2 ο T1-1 T2 T1, T2: transformations that fit the (same) kernel to each impression. Bring top kernel onto the bottom kernel Overlapped impressions. Green: common ridges. w05-fingerprints Biometrics - Summer

56 Mosaicking Minutiae matching w05-fingerprints Biometrics - Summer

57 Mosaicking Performance w05-fingerprints Biometrics - Summer

58 Company(web site) Sensor FAR(%) FRR(%) Biolink USA (biolinkusa.com) BiometricId (biometricid.com) Performance Claims by Vendors Optical Optical Startek (startek.com.tw) Optical IOSoftware (iosoftware.com) Optical Identix (identix.com) Optical NEC (nectech.com) Solid-State 2x10-4 5x 10-2 Biometrix Int (biometrix.at) Solid-State Pollex (pollex.ch) Solid-State Sony (sony.com) Solid-State Performance discrepancy between lab systems and deployed systems is due to difference in acquisition conditions w05-fingerprints Biometrics - Summer

59 Application: Credit Card A Touch of Money [FP4] The credit card has a sensor which will scan, extract features from a finger Need to register your finger print to the credit card company When you present your card to the point-of-sale terminal, secure communication will be established Verification of the validity of the card will be done You have to present your finger to the sensor on the card. Verification of your finger print with that from the data base will be performed. If verified, a digital signature will be sent to the point-of-sale terminal. You will be charged with the purchase. Only the vendor and your account information will be sent to the credit card company. Your finger print remains in the card. w05-fingerprints Biometrics - Summer