Classifications

• • G—PHYSICS
  • G07—CHECKING-DEVICES
  • G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
  • G07C9/00—Individual registration on entry or exit
  • G07C9/30—Individual registration on entry or exit not involving the use of a pass
  • G07C9/32—Individual registration on entry or exit not involving the use of a pass in combination with an identity check
  • G07C9/37—Individual registration on entry or exit not involving the use of a pass in combination with an identity check using biometric data, e.g. fingerprints, iris scans or voice recognition
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F18/00—Pattern recognition
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
  • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
  • G06V40/12—Fingerprints or palmprints

Definitions

• This invention relates in general to an automatic computer controlled technique for matching a set of points in a reference file record (hereinafter "File") against a set of points in a subject or search record (hereinafter "Search"), wherein each of the File and Search points have values associated with certain predetermined parameters. More particularly, this invention relates to a fingerprint matcher in which the positional parameters of a set of minutia points in a File are compared against the positional parameters of a set of Search minutia points to deter ⁇ mine whether or not it sufficiently matches so that the Search subject can be verified..
• File reference file record
• Search search record
• Fingerprint matcher mechanisms are known.
• One such Matcher is described in Wegstein, J.H., The M-40 Fingerprint Matcher, National Bureau of Standards Technical Note 878, US Government Printing Office, Washington, DC 1975.
• the matcher mechanism described in the Wegstein reference employs the position of each File minutia and Search minutia in X and Y coordinates and. the angular value in degrees.
• each File point is compared with each Search point, in each of these X,Y and angle parameters so as to provide displacement values for a difference table.
• a difference table is a table in which each entry is a comparison of a Search point against a File point.
• the table includes the displacement between the ' two points in the X direction, the displacement between the two points in the Y direction and may also include the angular displacement between the two minutia points being compared.
• a plot of the X displacement values and the Y displacement values creates a dif ⁇ ference plane.
• Certain clusters of points in the difference table are used to provide an indication of the extent to which the Search fingerprint might be displaced or rotated relative to the fingerprint from which the set of File points was obtained. Certain figures of merit or matching scores are obtained. Only if the matching score exceeds certain thresholds, is the. Search finger ⁇ print deemed to correspond with the File fingerprint.
• a type I error is the failure to verify a match when the subject finger is indeed the same as the finger from which the File was obtained. This is an incorrect rejection in an access system.
• a type II error is the verification of a subject fingerprint which in fact is not the same as the fingerprint from which the File was made. This is an incorrect admission in an access system.
• va-lue of any matcher mechanism is the extent to which the matcher contributes to increasing the speed of operation of a verification system, to reducing type I errors and to reducing type II errors.
• the Matcher device of this invention in ⁇ volves a number of major components which operate together to provide the total decision making in iden ⁇ tification that is appropriately termed matching. More parti ⁇ ularily, the Matcher has four separately scaled comparator segments which are integrated "so as to provide a variable scale matching function. The Matcher also includes a scorer mechanism. However for the Matcher to operate to provide decision making, it must also have a storage for the File minutia points and a storage for the Search minutia points. It is a combination of all of these components that constitutes the.Matcher.
• the difference table includes X differential values and Y differential values.
• the difference plane is one on which the X differential and Y dif ⁇ ferential values of the difference table are plotted.
• the difference table (D.T. herein) and the difference plane have to be kept in mind as distinct from the image plane on which the minutia points appear.
• an entry on the difference plane is a point (a) whose X coordinate represents the difference along the X axis of the image plane between a given file point and a given search point and (b) whose Y coordinate represents the difference along the Y axis of the image plane between that file point and search point.
• Q is used in-different contexts with different sub-scripts to generally refer to how tightly various points are bound to one another. It is a measurement of relative proximity. Is is roughly analogous to spatial peaking, with reference to space on the difference plane not on the image plane.
• Q ⁇ will refer to a measure of how close ,two points on the difference plane are to one another.
• Qs will provide a measure of how close a par ⁇ ticular point on the difference plane is to all the rest of the points in the difference plane.
• the measurement Qt will provide an indication of how close to each other are all of the points on the difference plane. But generically this measurement of proximity on the difference plane will be referred to as Q.
• QT is used herein, particularly in the FIG's., interchangeably with Qt to mean the same thing.
• the minutia referred to herein are the clas ⁇ sical minutia, namely line endings and line bifur ⁇ cations. These are the known type of minutia which are traditionally used to identify a fingerprint.
• the location of each minutia on the image plane is indicated in X and Y coordinates and is represented by the picture element (pixel) which contains the end point of the minutia.
• the angle of the minutia in the image plane is also employed to identify the minutia and can be employed as part of the matching mechanism.
• the embodiment of the invention described herein em ⁇ ploys that angle only in terms of an initial qual ⁇ ification criteria for entry on the difference table. The angle is not thereafter used in the embodiment described although the matcher mechanism can be gen ⁇ eralized to include angle differential values in the difference table as a contributor to the Q factor.
• this invention involves a fingerprint identification system in which a set of Search identification points derived from an optical scan of a finger are compared against a set of File identification points for the same finger of the individual involved. These identification points are individual points in an X and Y plane that represent stable minutia.
• the matcher system of this invention divides
• a technique of providing a correction factor based on a difference table analysis is employed.
• the Search points and File points are compared to provide an initial (first) difference table based on the distances between the points being compared in the X axis and Y axis.
• a proximity value Q is calculated for each entry on the difference table. 3ased on these proximity values, the initial difference table is purged of all but one entry representing each Search point. The higher Q value points are the ones retained.
• This much reduced purged (second) difference table is then used as a basis for recalculating a proximity value for each entry on that purged table.
• the entry on the purged table with the highest proximity value Q is treated as a point of origin on the difference plane in that the DX and DY values are reduced to zero.
• a comparable translation is made for the other points on the difference plane.
• This • treats the Search point/File point duo which constitutes the highest Q value on the difference table as if the two points where coincident.
• a cor ⁇ rection factor is calculated based on the average remaining difference plane spread between Search and File point pairs on the purged difference tab-Le..
• the . appropriate shift and correction factors are applied to the image plane position of the Search points represented on the purged difference table.
• a third difference table is established based on the corrected Search point positions.
• the third difference table is screened for difference table entries representing Search and File points that remain further apart than a predetermined number of pixels and is also limited to a predetermined number of entries based on those having the highest proximity values in the purged difference table.
• This now more limited (fourth) difference table is used as a basis for recalculating the proximity values for each entry therein. These recalculated proximity values are used as a basis for scoring the comparison to determine identification.
• the scoring technique also employs parameters that include the number of entries on the limited difference table, the number of entries on the purged difference table, the proximity score of the entire purged difference table, and a total segment score which is based on a ratio of the number of lines in the limited difference table and the number of lines in the purged difference table.
• FIG. 1 is a block diagram illustrating the relationship of the matcher mechanism of this invention to the equipment that provides the set of File minutia points and Search minutia points to be compared in the matcher, which matcher provides a Q value, other figures of merit and an indication of verification.
• FIG. 2 is a block diagram illustrating the processing of the matcher of this invention.
• FIGs. 3A through 3F is a- diagram that is primarily a flow chart illustrating in greater detail i i the comparator and correction processing set forth in i
• FIG. 3G I FIG., 2. To facilitate understanding the transition i from one of these components to the others, FIG. 3G
• FIGS. 3A through 3F shown in each of the FIGS. 3A through 3F.
• FIGs. 3H and 31 show a varient on the FIG.3A-3F embodiment in which correction factors are provided in the Y axis as well as in the X axis.
• FIGs. 3H and 31 correspond to FIGs. 3E and 3F respectively *
• FIG. 3J illustrates four points in the image plane. These four points are the File Point and Search Point pair S' and F' which has the highest Q value on the difference table. .
• Another two points illustrated are a difference table pair Sk and Fk which are being compared to the pair S 1 and F 1 respectively for the purpose of calculating the scale factor components SXT and WYT as indicated at operating box 38a of FIG.3H.
• FIG. 4A is a block diagram illustrating the four segment arrangement of this invention wherein the set of File and Search points are divided into four segments to provide four separate parallel stages of comparator and correction processing.
• FIG. 4B is a schematic/diagramatic illustration of a sixteen segment arrangement of this invention.
• FIG. 4B illustrates a two dimensional segmentation and indicates the degree of overlap of the File segments. There is no overlap of the Search segments.
• FIGS. 5A and 5B are flow charts indicating scoring techniques which are employed in connection with the FIG. 2 comparator to provide ID verification.
• FIG. 5C is a table of the Q values of a perfect match as a function of difference table line entries.
• FIGS. 6A through 6C illustrate a variant on the FIG. 3 embodiment in which two primary modifications are made.
• a first modification is that the FIG. 3D purge of the initial difference table is modified to assure that in the purged difference table there will be one line entry for each Search point IS
• FIG. 6A can be compared with FIG. 3D for an understanding of the changes involved.
• FIG. 6 shows the additional correction factor SIV.
• FIG. 6B shows deletion of the correction factors SIW and COW.
• FIG. 6C shows the correction of the X values of Search points IS by SIV as well as by COV.
• FIGs. 7 and 8 are flow charts illustrating enrollment techniques which are preferred for use in generating the File points to be held in the store 18,
• FIGs. 9 and 10 illustrate one known optical scan technique for generating the fingerprint image from which ..File and Search points are extracted.
• the matcher mechanism provides a score that indicates how well the set of minutia of a subject fingerprint image (the Search points) matches a re ⁇ ference set of minutia (the File points).
• a subject fingerprint image may differ from the reference fingerprint image even though both are taken from the same finger of the same individual. Displacement and rotation of the finger causes some of these differences. Differences are also due to the plasticity of the finger which results in some change in the configuration of the minutia points as a func ⁇ tion of changes in the amount of force and the direc ⁇ tion of the force applied by the individual when plac ⁇ ing his or her finger on a platen or other support.
• the condition of the finger varies from day to day and is a function of recent environment and use. The result is that the set of Search minutia points from a subject imagt differ from the set of File points in a reference file, even though the two fingerprints are from the same finger.
• Each minutia has X and Y coordinates on the image plane.
• the Matcher of this invention centers around the establishment of a difference table re ⁇ lating to these X and Y image plane position par ⁇ ameters.
• the two sets of minutia points to be com ⁇ pared, the File set JS and the Search set IS, may be derived from an image produced by an optical scan of a fingerprint such as that disclosed in United States Patent Number 4,322,163 issued March 30, 1982 and entitled FINGER IDENTIFICATION.
• the fingerprint image thus provided may be a binary image composed of a number of picture elements (pixels) that permits iden ⁇ tification of the standard minutia, specifically the line endings and line bifurcations.
• the minutia extracted from such an optical image may be obtained by any of a number of known techniques. Such techniques are described in Stock, R.M., Automatic Fingerprint Reading Proceedings, of the 1972 Carnahan Conference on Electronic Crime Counter- measures, University of Kentucky, Lexington Kentucky, " 1972, p.16-28 and Banner, C.S., and Stock, R.M_, FINDER The FBI's Approach to Automatic Fingerprint Identification, Proceedings of a Conference on -the Science of Fingerprints, Home Office, London, England, 1974.
• the minutia extracted are stored as a re ⁇ ference record (the File) or as a subject record (the Search points) in memory units of the automatic clas ⁇ sification- and identification equipment. Each record is a set of points. Each point has X and Y positional coordinates and an angular coordinate. Thus each point is represented by three parameters.
• the matcher mechanism described herein treats each such minutia point as a point and makes no distinction between a line ending minutia point and a bifurcation minutia point.
• the set of minutia points involved may be extracted from the image by any one of a number of mechanisms. Indeed, the matcher mechanism of this invention does not require that the set of points involved be minutia points. All that is re ⁇ quired is that the set of points is deemed to be uni ⁇ que to the fingerprint of the individual involved so that a unique -identification may be provided. Ac ⁇ cordingly, it should be understood herein with respect to the disclosure and the scope of the claims, that reference to identification points is not necessarily limited to minutia identification points.
• the System FIG . 1
• FIG. 1 illustrates the re ⁇ lationship of the Matcher 10 of this invention to the equipment with which it may be associated.
• the optical scan mechanism 12 such as that described in the re ⁇ ference Patent No. 4,322,163 and shown in more detail in FIGs. 9 and 10, provides a binary image.
• the iden ⁇ tification points extraction mechanism 14 extracts from that image a set of points that identify the fingerprint involved.
• the extraction mechanism 14 may be one of those described in the minutia extraction references identified above.
• the set of subject points are maintained in a Search point memory store 16 so that they can be compared by the matcher 10 with a set of points in a File point memory store 18.
• the matcher 10 compares the contents of the two memory units 16 and 18 in the manner described hereinafter to come up with a score which indicates verification.
• the set of points in the File store 13 may be de ⁇ termined in any fashion. However, for purpose of minimizing both the type I and type II errors, it is preferable that the set of points in.the File store 18 be derived from the same mechanism which derives the set of points in the Search store 16. -
• the File points JS and Search points IS are stored in memory units 16 and 18 by three parameters, namely the X value, Y value and angle value on the image plane.
• both JS and IS points are stored in Y value order.
• JS_ has the lowest Y value of any File point
• JS2 has the next Y value, and so forth.
• FIG. 2 An important aspect of the system of this invention is the segmenting of the fingerprint image into a number of separate zones and the application of the comparison techniques and matching techniques to each zone or segment separately. This multiple segment arrangement is described in connection with FIGs. 4A, and 4B.
• the scoring techniques of FIGs. 5A and 5B use the quality scores from each segment of the matcher.
• FIGs. 2 and 3A through 3J The Comparator (FIG. 2)
• FIG. 2 An understanding of the ' block and function diagram of FIG. 2 will facilitate following and under ⁇ standing the more detailed flow chart arrangement shown in FIGs. 3A through 3J.
• the points in the stores 16 and 18 are compared and screened at 20 by a known comparison technique so as to provide the known type of difference table 22.
• the Qs value for each entry on the difference table is calculated at 24 and then the difference table is rearranged so that it is ordered in sequence by the value of Qs at 26. This means that the first entry in the difference table is the one with the highest Qs value.
• the routine taken at 28 serves to purge from the difference table all but one entry representing a Search point. That is done by selecting the entry having the highest Qs* value for each Search point.
• a comparable routine is then performed at 30 to purge the difference table of multiple File point entries by retaining only that entry having the highest Qs which represents the File point involved.
• this purged difference table is now employed as the difference table for the system and the Q s values are recalculated based solely on the entries in the purged difference table.
• the purged difference table is reordered so that the high ⁇ est Q value is first as indicated at 36.
• the image plane X and Y values for the Search points are shifted to make the Search and File point pair S 1 , F' for the highest Qs entry on the purged difference table coincident.
• This shift is made by making the DX and DY values for that first entry S 1 , F 1 on the reordered purged difference table equal to zero.
• the same shift is made to all the other DX and DY values thereby effectively shifting all of the Search points Sk by the same amount.
• the Search points are further relocated based on the scale factor. Speci ⁇ fically, after the above shift of positions, the DX and DY values in the reordered purged difference table are corrected by the scale factor. It should be noted that in the simple embodiment of FIGs. 3E..and 3F, only the DX values are corrected by the scale factor. - The scale factor operates so that the larger the value for DX or DY, the greater will be the magnitude of the relocating correction. Since the value for DX and DY has been set to zero at stage 40 for the difference table entry having the highest Qs, the search point for that highest Qs entry will not be affected by the scale factor correction.
• the re ⁇ location of the Search points by shifting and by scale correction is effected by changing the DX and DY values in the difference table rather than by af ⁇ fecting the image plane position values in the store 18.
• the number of entries on the limited difference table is limited to fifteen, which are the fifteen entries on the purged difference table that have the highest Qs value.
• these entries on the limited difference table differ from the corresponding entries on the purged difference table because the values of DX and DY have been changed to reflect the corrected and relocated Search point positions.
• These new values for DX and DY in the limited difference table are used as the basis for calculating the Qc and Qs values for each of the up to fifteen points in the limited difference table.
• the correction routine serves to shift and correct the Search points so that when the correct finger is presented for identification, the Search points are repositioned to be more in line with the File points and when the incorrect finger is presented the Search points tend to be shifted more out of line with the File points than they would be without the correction routine. Accordingly, the correction routine of this invention reduces.,both type I and type II errors.
• the Comparator Flow Chart (FIG.3A-Through 3G)
• An initial difference table is created based on a comparison of the set of File points in the store 18 and the set of Search points from the subject image in the store 16. Within certain limits, each File point JS is compared with each Search point IS in establishing the table. No distinction is made bet ⁇ ween line ending points and bifurcation poi ts. Each minutia, is treated as a point. A portion of such a difference table is illustrated in FIG.3B.
• Another entry DY is made equal to the dis ⁇ placement along the Y axis.
• a third entry may ' _.-; made indicating the angular difference between the two points.
• the DX and DY entries are the number of pixels between the two points. A plot of all DX and DY entries creates the difference plane.
• the number of entries in the initial dif ⁇ ference table (D.T.) is restricted in that for a given File point certain Search points are not included.
• the Search points IS that are excluded from the difference table (D.T. ) comparison with a given reference point are where either (i) the Search point is outside of a 61 x 61 pixel box centered on the File point; or (ii) where the angle of the Search point is outside of a band of plus or minus 22.5 degrees around the angle for the File point.
• the angle values for each point are used only for this initial screening. They are not otherwise used, al ⁇ though the device of this invention could be extended to use angle information as a further indicator of matching. In this fashion, the number of entries in the initial difference table is limited.
• FIG. 3B is a representation of a portion of a typical difference table. Each entry is given a line number. Each File point JS is compared with each Search point IS which is within the 61 x 61 pixel box centered on the File point. The X axis and Y axis displacements, DX and DY respectively, are entered in the difference table. The difference table is ordered by the Y value of the reference points. Because of the ordering of the JS and IS points in the memories 16 and 18 it is pos- sible, to include a loop routine which speeds up the comparison. Any Search point IS with a Y value below that of the window set up on a given JS, need not be compared with any subsequent JS points. Thus a deter ⁇ mination that the Y value of an IS is below the window means that the next IS point can be immediatel " re ⁇ viewed.
• FIG. 3B indicates the routine used to cal ⁇ culate the Q value for each entry on the difference table.
• the Q value for an entry. is the value Qs. That value Qs is the sum of individual values Qc.
• Each individual value Qc is based on a comparison of a given difference table entry with one other difference table entry.
• a high Qc indicates that the two entries- are relatively close to one another on the difference plane.
• the two difference plane points are compared by taking the- difference between two DX values and adding that to the difference between the two DY values to obtain the TR value, as shown by the equation 24a in .FIG. 3B.
• This TR value is then sub- stracted from a constant KR, which has the value 15 in this embodiment, to come up with the Qc value.
• FIG. 3C shows the routine for ordering the ⁇ itterence table in terms of the Qs values assigned to eacn entry on the difference table.
• the ordering rou ⁇ tine is a fairly standard routine.
• the ⁇ function 50a simply means that the out ⁇ put value SQ to the scoring system is zero when there is only one line in the difference table. That is, if ST is equal to or less than "one", then SQ is im ⁇ mediately put out as having the value zero. This will provide an indication to the scoring system of a mis ⁇ match and identification will be negatived.
• the value KSW increments to "one" after the functions performed by the blocks 38 through 44 of FIG.
• the purge reduces the number of random correlations to further improve type II error.
• the number of entries S 0 i£ in the purged difference table is retained as a value to be employed in the cal ⁇ culation of the output factor SQ.
• the portion of the FIG. 3D that is enclosed in dotted lines and indicated by the reference numeral 31 is a specific purge routine which is repeated at other stages of the processing of the various embodiments described herein.
• the purge routine 31 portion of the FIG. 3D purge technique is a routine in which a purge buffer P has a single slot for each line entry on the D.T.
• the purge buffer is initialized to zero.
• a "zero" status indicates that the corresponding line entry in the D.T. is to be retained.
• the purge routine 31 is undertaken to delete that line entry.
• three indicies are employed.
• the index "i" indicates the line on the D.T. being examined.
• the index "c” indicates how many lines are to be purged out.
• the index "s” indicates where a retained line entry is on the D.T. after purge of one or more other lines.
• the count c of line entries to be eliminated is incremented by one. as indicated at 3In and that line entry is placed at the line 256 as indicated at 31p. Placing a line entry at line 256 effectively eliminates that line from the D.T. since the number of lines ST on the D.T. is always substantially fewer than 256 and thus that line will be eliminated at the function box 322.
• the purge routines 31b, 31c, 31d, 31e and 31f indicated in other FIGs. are identical in function to the routine shown at 31 in FIG. 5D. Those other purge routines operate on D.T. line entries which are deleted for other reasons.
• the purged difference table then goes through the Qs calculation procedure of FIG. 3B and then is ordered by the new Qs values as shown in FIG. 3C.
• the difference table entries are provided at D as a basis for the calculation of a scale factor CO-7 as indicated in FIG. 3E.
• the Search point represented by the highest Qs entry on the purged difference table is deemed to be co.incident with the corresponding File point represented by that same entry on the difference table. That is, they are deemed to be coincident on the image plane. With respect to each other entry on the difference table. all other Search points are deemed to be displaced from the corresponding reference point because of distortion. A correction factor is calculated to reflect that distortion.
• a correction factor is calculated to reflect that distortion.
• a separate scale factor component SXT is calculated for each Search point, there being a separate Search point for each entry on the purged difference table.
• the scale factor component SXT for the highest Qs entry on the difference table is zero.
• FIG. 3H discloses an embodiment where a scale factor correction is made in both directions. Observation shows that in certain types of fingerprint scans the Y distortion is relatively small compared to the X axis distortion and in such cases the FIG. 3E embodiment is sufficient.
• the scale factor correction SXT for the individual minutia SK, i involved is based on the displacement value XIJ.
• the individual scale factor correction SXT is based on the assumption that the highest Q search point S 1 has been shifted to coincidence with the file point F' with which it is paired as an entry on the difference table.
• Each other search point Sk is shifted on the image plane the same amount. If the search image does indeed correspond to the file image, then the shift normally moves these other Search points- Sk closer to the corresponding File points Fk with which they are paired as an entry in the difference table. To the extent that the other Search points Sk are not coincident with the corresponding File point Fk, that lack of coincidence (for corresponding search and file images) is presumed to be due to plastic distortion in the fingerprint image.
• the correction factor component SXT for the correction factor GOV has to be normalized for the distance that Gk is from S 1 .
• the correction factor component SXT is the ratio between (a) the distance XIJ between the two points Sk and Fk being compared and (b) the distance XI that the search point Sk is from the reference point S 1 . This is shown in function box 38a.
• the actual correction factor COV which is shown in box 38b, is the difference between the value "one" (which represents no correction) and the average of the individual correction factor components SXT.
• the average of the individual correction factors is simply the sum of the individual correction factors SXT (which sum is represented by the symbol SX in Fig. 3E) and the number of search points involved ISX.. If the two points being compared Sk and Fk are coincident, the XIJ value is zero and the component SXT for the correction factor COV is zero. The difference between the average correcting factor and the value "one" is the actual correction factor COV that is applied to the X values of those search points on the image plane.
• the screening factor DD at 38d is less than eight pixels, then the correction component SXT is not calculated. This tends to eliminate over correction due to a noise point, it being unlikely that a good Search point would be that close to the Search point of highest Q.
• the set of Search points represented by the purged difference table can now be repositioned on the image plane by (a) first translating each Search point by an amount equal to that translation which brings the highest Q Search point coincident with its companion File point and then (b) by relocating Search points based on the scale factor COV.
• FIG. 3F illustrates the routine by which the repositioning of each search point Sk occurs.
• the formulas shown in the function box 40a and 42a achieve the result of both translating the search point and adjusting it in accordance with the scale factor correction COV.
• An understanding of the function iioxes 40a and 42a is facilitated by reference to the sample points on an image plane as shown in FIG. 3J.
• the distances X and Y are calculated as indicated in block 40a and as shown in FIG. 3J.
• the notation XA represents the distances XI referred to in FIGs. 3E and 3J.
• the location of the point Sk,i on the image plane is based at the distances XI and YI from the file point F 1 .
• F' is from the difference table entry having the highest Q.
• the search point S 1 is positioned coincident with the file point F'.
• the output from the FIG. 3B function is applied so as to permit calcu ⁇ lation of the SQ value.
• the SQ value is essentially the sum of the Qs values from the limited difference table (QT), times the number of entries on that table (ST), divided by the number of entries in the purged difference table (Sold)- ⁇ l ⁇ e value KRR is to normalize the value of QT since each Qc is a double contribution and the mul ⁇ tiplication of QT by 0.5 normalizes the result. It should be noted that the value for ST cannot be greater than fifteen whereas the value of the de ⁇ nominator S 0 i c j will normally be greater than St.
• the minutia point comparison mechanism importantly includes the combination of (a) the purge of the difference table so as to limit it to only the more likely Search and File point comparisons and (b) the technique of shifting and correcting the effective position of the Search points on the image plane before calculating the proximity values which are used to determine whether or not a match has been made.
• This combination tends to result in a very significant improvement over other matcher designs in type I error. But it also improves type II error.
• the features of this matcher design have the advantage of moving the trade off between the different types of errors and the speed of operation to an improved level because the nature of the mechanism improves both types of errors.
• FIGs. 3H and 31 are comparable to FIGs-. 3E and 3F excep't that FIGs. 3H and ' 31 illustrate the steps required to provide scale factor correction along two axes. Accordingly, where the functions are essentially the same, the same reference numberals are used in FIGs. 3H and 31 as are used in FIGs. 3E and 3F. However, some of the description is repeated herein to facilitate an understanding of this two axis correction.
• the distance XI is a distance on the image plane between (a) the X value of the search point S' that is a member of the pair having the highest Q in the difference table and (b) the X value of the search point Sk being evaluated. These two distances are respectively the distances XXU and SXK.
• the Y axis distance YI is completely analogous.
• the initial parameters for the scale factor correction calculation shown in FIG. 3H are as indicated in the block 39g and the blocks upstream from 39g.
• the X and Y values on the image plane for the Search point S' and file point F' are those values XXU, YYU, XXF and YYF indicated in 39g.
• FIG. 3J an example on the image plane of the Search point S' and File point F' from the top line in the difference table and an example of one other Search poink Sk and File point Fk from some other line on the difference table.
• function box 39f Providing that the Search point Sk is outside the eight pixel proximity criteria of function boxes 38d and 39d., the values shown in function box 39f are provided to permit calculating the scale factor components as shown in either function box 39m or 39n. Which of these scale factor components are calculated is determined by the comparison shown in decision box 39a.
• the decision box 39r is significant only during enrollment. It is an enrollment flag. During the enrollment procedure described hereinafter the deletion routine provided by the buffer 45 is not undertaken. But for immediate purposes this enrollment flag 39r can be ignored.
• the appropriate correcting scale factors COV, SIW, COW, SIV are set to values of one and zero as appropriate and as indicated in the two function boxes 39y.
• FIG. 31 is quite similar to FIG. 3F.
• the shift at operator box 40a is the same.
• the relocation based on scale factor correction is, as indicated at boxes 42b, 42c performed in both the X and Y axis.
• the decision box 42d determines whether the scale factor corrections to be applied are the COV/SIW pair where X axis distortion is greater than or equal to Y axis distortions, or the SIV/COW pair where Y distortion is greater than X axis distortion.
• the purge routine 31d is to eliminate those line entries where the File point Fi is outside the 43a window around the corresponding Search point Si.
• FIG. 6A, 6B and 6C illustrate a variation on ⁇ the embodiment illustrated in FIGS. 3A through 3F. This is a presently preferred embodiment which appears to provide some enhancement over the FIG. 3A-3F embodiment.
• FIGS. 6A, 6B and 6C correspond to FIGS. 3D, 3E and 3F, respectively. Accordingly, it should be understood that the disclosure relative to FIGS. 3A, 3B, 3C and 3G substantially apply to this FIG. 6 embodiment except for a couple of details mentioned below.
• KR is set at the value of 7. When this is done, it has been found unnecessary to include the decision box TR:KHR.
• FIG. 6A illustrates a presently preferred purge technique for the difference table (D.T.).
• This technique differs from the technique illustrated in FIG. 3D primarily in that the FIG. 6A purge incorporates what is in effect an "add-back" feature that takes into account the fact that the purge of all multiple File and multiple Search points (the IS and JS points) may result in a purged D.T. having no entry for certain Search points that are represented in the initial D.T. It has been found to enhance accuracy to include an entry in the purged difference table for each IS point that is incorporated in the initial D.T. Accordingly, after the Search and File purge routines have been followed, it is desirable to add back the highest Q score D.T.
• the result is a difference table having a net purge such that there are no more line entries than Search points and no more than one line entry for each Search point. There may be fewer line entries than Search points because of the initial qualification " routine so that any Search points which are not represented in the initial difference table are not represented in the purged difference table.
• a first buffer MS is referred to as a mark search buffer.
• the MS buffer has a slot for each IS point that is a member of an entry on the initial D.T.
• the slot in the MS buffer is used to indicate whether or not that IS point is retained as a member of some entry on the purged D.T. after the raw purge.
• this MS buffer has a "zero" designation for each IS point.
• the "zero" designation means that the IS point is not represe ted in the purged D.T.
• the MS buffer entry for the corresponding IS point is changed to the designation "one" indicating that the IS point is to be represented in the purged D.T. by one of the entries from the initial D.T.
• the mark file buffer MF has exactly the same structure and performs exactly the same function for each of the file points JS.
• an array of information is provided that permits a routine of determining which IS points are not represented after the purge routine has been completed. Then an appropriate line entry from the original D.T. can be selected so as to include a representation of each IS point in the purged D.T.
• the third buffer P is called a purge buffer. It has a slot for each line entry on the initial D.T.
• the P buffer simply contains an indication as to whether or not the corresponding line entry on the D.T. is to be retained or eliminated from the purged D.T.
• a "zero" designation in a P buffer slot means that the corresponding line in the initial ordered D.T. is to be retained in the purged D.T.
• the designation "one" in the P buffer slot means the corresponding line in the initial ordered D.T. is to be eliminated from the purged D.T.
• the use of the "zero" and “one” status designations in the P buffer thus differ from the use of those status designations in the MS and MF buffers for reasons that have to do with the advantage of initializing all slots in each of the three buffers to zero.
• the operations at 53a, 53b and 53c all indicate initializing the three buffers involved to a zero status in each slot of each buffer.
• the number of slots in the P buffer is equal to the number of lines (St) in the initial D.T.
• the number of slots in the MS buffer is equal to the number Set of Search points and as indicated at 5If the number of slots in the MF buffer is equal to the number Fct of File points.
• the routine generally shown at 54 in FIG. 6A illustrates the raw purge routine.
• Each of -the "St" .. line entries in the initial ordered D.T. are compared at 54a with the status indicated for that IS. point in the MS buffer.
• the line entries are so processed in order of Qs value. If that IS slot is zero, then that D.T. entry is compared, as indicated at 54b, with the status of the JS point in the MF buffer. If that status is also zero, then as indicated by the function box 54c, the status of the IS point and JS point in that line entry is changed to "one" in the MS and MF buffers respectively.
• This routine is repeated for each of the entries on the initial ordered D.T.
• the P buffer will hold a designation "zero" for a- set of D.T. line entries in which there are no duplicate IS points and no duplicate JS points.
• the P buffer will designate all other D.T. line entries as "one".
• the routine shown at 55 adds into the P buffer a "zero" status (a retain designation) for D.T. line entries having an IS member which is not represented after the raw purge indicated by the routine at 54.
• each MS buffer slot with the status "zero" meaning that IS is not represented after the raw purge at 54
• each D.T. entry by Qs order is compared to identify a D.T. entry having the IS missing from the MS buffer as a member.
• the P buffer slot corresponding to the identified D.T. entry has its status changed to "zero" and the slot in the MS buffer designating the IS member of that line entry has its status changed to "one".
• the P buffer is caused to designate at-least one, but no more than one, line entry from the initial D.T. having each IS member in the initial D.T. Because the FIG. 6A purge routine operates on an initial D.T. ordered by proximity valve Qs, the highest Qs for each IS is the line on the purged D.T.
• the purge routine 31e like comparable routines shown in FIGS. 3D, 3 ⁇ , 3F, 3H and 31 provides for the actual purge of all line entries where the P buffer has the status "one".
• FIG. 6B can be understood as adding the CIV correction factor to the FIG. 3E routine or. as deleting the COW and SIW correction factor routines from the FIG. 3H arrangement.
• the SXT value from each Search point IS is calculated as in
• FIG. 6B the criteria shown at 38d in FIG. 3E is not employed and thus those Search points having relatively small XI values are used in the FIG. 6 embodiment.
• the WXT values and corresponding SIV correction are calculated in the FIG. 6B embodiment:.
• the FIG. 6B arrangment can be thought of as taking the X-axis distance between Sk and Fk (see FIG. 3J) , which X-axis distance is XIJ (i.e., XI-XJ) as shown at 56c and normalizing that presumed error for both X and Y distortion.
• XIJ XIJ
• XI-XJ XI-XJ
• dividing the "error" XIJ by the X distance (XI) between S' and Sk normalizes XIJ for distortion along the X-axis (see 56d) .
• dividing XIJ by the Y distance (YI) between S 1 and Sk normalizes XIJ for Y-axis distortion.
• the SXT values are summed and the WXT values are summed.
• the average values are used to provide COV and SIV as shown at 56f. The result is that COV and SIV both provide a measure of X-axis "
• the purge routine 3If in FIG. 6B is undertaken to eliminate from the D.T., those line entries which result in a correction factor component SXT or WXT greater than 20 percent. As indicated at 56a and 56b, if either of those components is greater than 0.2, then the purge buffer 58 has its slot that corresponds to that line entry given the status "one". The purge routine 3If operates exactly as does the purge routine 31 in FIG. 3D to eliminate that line entry and to bring all line entries that are below the eliminated line entry up one line each.
• the shift correction shown in FIG. 6C at 40a is the same as in the FIG. 3F embodiment.
• the correction in the X value of Si takes into account both COV and SIV factors. In order to avoid overcorrecting, the value of the correction due to each factor is divided by two.
• a window is created around the shifted and corrected Si point. This window is created on the image plane.
• the window extends 5 pixels in both directions along the X-axis from the Si point and 20 pixels in both directions • along the Y-axis from the Si point. If the corresponding file point Fi falls outside that window, then the corresponding line entry is dropped from the difference table by the technique shown in FIG. 3F.
• the finger ⁇ print or finger image is divided into four segments. These are four vertical segments.
• the X axis is divided into four segments and the zone associated with a segment extends along the entire Y axis.
• the Search point stores do not overlap.
• each Search segment is compared against a corresponding File segment in a separate comparator 17a through 17d, which comparator is essentially the device illustrated in FIGs. 2 and 3. Specifically all of the units shown in FIG. 2 from the compare and screen unit 20 through the calculate unit 50 are replicated for each of these four comparators 17a through 17d.
• each of the FIG. 2 units 17a through 17d provide five items of output data as indicated and-defined in FIG. 4A. It is this output data which is processed by the scoring system of FIGs. 5A and 5B to provide verification of iden ⁇ tification.
• the four segments illustrated in FIG. 4A are vertical segments which break the image up along the X axis into four zones which overlap along the X axis. Because of the overlap, each segment covers somewhat more than one third of the image.
• FIG. 4B illustrates an extension of the FIG.
• FIG. 4A embodiment substantially apply to this FIG. 4B embodiment except there are now sixteen comparators 17 employed in the FIG. 4B embodiment. There are, accordingly, sixteen sets of output data, one set from each comparator and thus one set representing each of the sixteen sub-spaces. This output data is processed by the scoring system of FIGs. 5A and 5B to provide verification of identification..-
• a fingerprint card provides the set of Search points that are to be compared with the set of File points
• somewhat greater flexibility with respect to time is possible and. much greater control over alignment may be provided. Accordingly, in such a system, the amount of overlap between the multiple segments may be appreciably less or if alignment is adequately controlled, there need be no overlap.
• the separate, simple linear scale factor calculated for each segment makes it easier to identify a cluster out of a noisy environment.
• the distortion and the noise that may exist in a fingerprint file tends to make it difficult to identify a cluster.
• the ability of the simple scale factor correction to compensate for the distortion reduces the masking of the cluster.
• the segmentation is particularly useful to aid in identifying a finger ⁇ print in a noisy environment. - .
• An important aspect of the system of this invention is the segmenting of the fingerprint image into a number of separate zones and the application of the comparison techniques and matching techniques to each zone or segment separately. This multiple segment arrangement is described in connection with FIGs. 4A and 4B.
• the scoring techniques illustrated in FIGs. 5A, 5B, and 5C are scoring techniques which use the quality scores from each segment of the matcher.
• FIG. 5A is a flow chart of the operation of one of four scoring units. Each of the scoring units are coupled to separate outputs from the four separate comparator segment units 17a through 17d. Thus the particulars of the following identification technique is based on the four segment embodiment of FIG. 4A. If any one of the four sets of ouput parameters meets the verification critera set forth in FIG. 5A then identification is verified. This is a quick procedure when there is a high degree of correlation between the Search points and File points in any one of the four segments and avoids going through the FIG. 5B routine.
• the QT ratio is the ratio between QT and what QT would be if there were a perfect match between the Search and File points in the limited set coming out of the process of FIG. 3F.
• FIG. 56 shows the QT values for other perfect match situations .for other size difference tables.
• FIG. 5A there are a number of other identification paths involving others of the five parameter outputs which permit identification to be verified and those paths are self evident from FIG. 5A. What is important to recognize is that there are two main outputs other than identification being verified.
• One of the outputs involves applying the value "one" to the "vote" parameter V if the correlation between Search points and File points is less than sufficient to provide a verification of I.D. but is greater than certain " critera set forth in FIG. 5A. However, if those critera are not met, the output from FIG. 5A is in effect a value for V equal to "zero".
• an output from the FIG. 5A unit (being the value V equal either to "one or "zero") is applied to the FIG. 5B unit.
• the inputs to FIG. 5B are the outputs from each of the four FIG. 5A units presuming that the I.D. has not been identified in any one of those four FIG. 5A units.
• a combined final matcher segment score SQ is calculated that takes into account not just the individual matcher segment scores SQ n but also the mutual matcher segment scores SQ mfn determined by the relationships between the four matcher segments taken two at a time.
• the calculation of the individual matcher segment quality score SQ n is shown at the operating box 50b in FIG. 3C.
• the combined quality score SQ from the various segments of the matcher is calculated in a fashion that takes into account not only the quality score for each segment SQ n , but also the relationships between the quality scores of each segment.
• FIG. 5A and the ST values (see FIG. 4A) .
• the SQ m n value between any two segments if added to the SQ value ' for each of the two segments being compared equals the SQ value for the two segments combined and considered as one segment.
• the product of the two ST values provides the factor to represent the additional pairing that would be possible if the two segments were considered as a single segment.
• the product of ST for segment m and ST for segment n provides an appropriate factor that exactly makes up for what is lost by segmenting where there is a perfect match.
• the two proximity value ratios (QT ratio) provide the appropriate down grading of the SQ m n contribution.
• FIGs 5A and 5B may not be required. It might only be necessary to rank a series of quality scores SQ (as calculated in box 52b) so that the Files for the top predetermined few are pulled for further investigation. In such a situation, the Search record is compared against each of a predetermined number of Files, selected for some fingerprint type category, and those having the largest SQ values selected. In such an application, the I.D. thresholds shown in FIGs 5A and 5B would not be employed. Enrollment (FIGs 7 and 8)
• the value of the Matcher is greatly enhanced by having an appropriate and effective enrollment procedure so that the Pile points include a reasonably good number of repeatable classical minutia.
• the enrollment techniques shown in FIGs. 7 and 8 are of appreciable importance.
• the automatic enrollment scan technique of FIG. 7 will normally provide a good usable enrollment File
• the manual enrollment technique of FIG. 8 can be employed when it is desired to generate the enrollment File on the basis of an enrollment officer visually picking out points to provide the enrollment File.
• a third example is the possiblity of en ⁇ rolling a finger from two different individuals in the same File. Access would be had in a high security situation only if the Search file contains the minutia points from the two individuals presenting their re ⁇ spective fingers at the same time.
• a point in the initial en ⁇ rollment file is deemed to be matched in any one of the subsequent enrollment scans only if that initial enrollment file point appears in the limited dif - ference table developed from the output of FIG. 3F. Because no ID verification is involved, the subsequent calculation of values such as Q ⁇ need not be under ⁇ taken. However, processing in enrollment does re ⁇ plicate the processing in verification up to the point where ID verification steps are required.
• the limited file formed at 60c becomes the basic file which three additional scans of the finger being enrolled are made. These three additional scans can provide additional verification for each point in the enrollment file. As indicated at 60d.
• N 8 or more
• a maximum output file of 30 points is established based on those points having the highest number of verifications.
• a technique of manual enrollment can be employed where the enrolling officer has the expertise and know how to make a selection based on visual examination that permits the creation of a set of File points on one or two finger scans. This can be quicker than automatic enrollment and sometimes provides a better file than does automatic enrollment.
• a single scan is taken and the extracted points are presented to the en ⁇ rollment officer who goes through a point by point procedure of selecting or rej-ecting the extracted points. If the enrollment officer selects forty or more points, the enrollment procedure is ended and those forty or more points become the File. If the enrollment officer picks fewer than forty points, the enrollment officer can execute an additional scan to provide an additional set of points for review.
• the extracted points R are presented to the enrollment officer automatically on a one to one basis on the image.
• a first counter N ck provides an indication of how many points have been reviewed by the enrollment officer and a second counter N ves provides an indication of how many of these points have been selected by the enrollment officer for inclusion in the File.
• the point selection routine proceeds as long as the number of points that have been reviewed c j is less than the number of points R extracted in the first scan. As indicated at decision box 62a the selection routine terminates once the enrollment officer has selected forty points. If fewer than forty points have been selected, the routine continues as indicated at 62b, 62c and 62d with the enrollment officer picking points one at a time until the number of points reviewed exceeds R or the number of points selected is equal to forty.
• the enrollment officer can either decide to take a second scan or to terminate the process and settle for fewer than forty points in the file. If a second scan is taken, .the counter N. ⁇ indicating the number of points received has to be reset to "one" but the number of points selected counter N ves is not reset. On the second scan, the enrollment officer by use of judgement can select whatever additional points are desired until a total of forty have been selected between the two scans or until the enrollment officer decides to end the selection process and settle for less than forty points in the File.
• FIGs. 9 and 10 illustrate an optical scanning technique that may be employed to provide the basic image* and data from which the Search points are ex ⁇ tracted that are placed in the store 12.
• FIGs. 9 and 10 indicate a known mechanism for optically scanning a fingerprint to provide the modulated- light beam input to a CCD array 77. Since it is described in the issued Patent No. 4,322,163, the disclosure here need not be in great detail. Suffice it, therefore, to say that a beam of light such as may be provided by a laser 66 is appropriately collimated by lenses 68 and 70 to provide the inter ⁇ rogating beam 71.
• a substantially transparent platen 62 is provided as a base on which an individual finger F may be placed.
• the platen 72 is mounted in a movable carriage 74 which permits moving the finger across the interrogating beam 71.
• the pattern of ridges and valleys provided at the back surface of the platen 72 causes the reflected light beam 75 to be modulated with fingerprint information.
• a focusing lens 76 focuses the image carried by the reflected light beam onto a linear array 77 of photo responsive diodes.
• An encoder element 78 which is affixed to the carriage 74 responds to movement of carriage 76 to produce a synchronizing signal each time the carriage moves a predetermined distance.
• the synchronizing signal causes the scanning circuit 80 to sequentially interrogate each of the photodiodes comprising the array 77.
• the output of the scanning circuit 80 is a train of pulses for each scan line. Each pulse represents a picture element or pixel.
• FIGs. 9 and 10 is but one example of the finger scanning techniques with which the matcher device of this invention may be used.
• the input to the array 77 can be the modulated light beam that is produced from a scan of a fingerprint card or the direct scan of a finger positioned in space without a platen.

Landscapes

• Engineering & Computer Science (AREA)
• Theoretical Computer Science (AREA)
• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Human Computer Interaction (AREA)
• Bioinformatics & Cheminformatics (AREA)
• Life Sciences & Earth Sciences (AREA)
• Artificial Intelligence (AREA)
• Multimedia (AREA)
• Bioinformatics & Computational Biology (AREA)
• Computer Vision & Pattern Recognition (AREA)
• Data Mining & Analysis (AREA)
• Evolutionary Biology (AREA)
• Evolutionary Computation (AREA)
• General Engineering & Computer Science (AREA)
• Collating Specific Patterns (AREA)
• Image Analysis (AREA)

Applications Claiming Priority (4)

Publications (2)

Family

ID=27109987

Family Applications (1)

Country Status (4)

Families Citing this family (2)

Family Cites Families (8)

• 1986
  • 1986-04-01 EP EP19860902617 patent/EP0216902A4/de not_active Withdrawn
  • 1986-04-01 WO PCT/US1986/000635 patent/WO1986005903A1/en not_active Ceased
  • 1986-04-01 KR KR860700856A patent/KR880700362A/ko not_active Withdrawn
  • 1986-12-01 FI FI864881A patent/FI864881A7/fi not_active IP Right Cessation

Also Published As

Similar Documents

Legal Events