MXPA00000492A - Method for determining an identification code from fingerprint images - Google Patents

Abstract

The invention relates to a method for determining an identification code from fingerprint images, wherein at least two of the following independent characteristics line spacing L, gradients G, curvatures K and bifurcations B are detected on the image area and a frequency distribution H is determined. On the basis of said frequency distribution, the characteristic values (mean value, variance, maximum value and classification value) and the characteristic values (Ci) of the selected bifurcations are determined, which form the vectorial components of the identification code C. The inventive method can be used to establish a short identification code which is relatively easy to determine and displays high recognition reliability for various applications.

Description

METHOD FOR DETERMINING AN IDENTIFICATION KEY FOR I IMAGES OF DACTILARTICAL FOOTPRINTS i Description of the Invention ji The invention relates to a method and a device for determining identification keys by i fingerprints or digital images of gray scale values respectively according to the generic terms of claims 1 and 25. They are for particularly valid for widely publicized forensic applications that require a very precise analysis of these characteristic details (position and type of detail as well as their orientation), which in turn requires a correspondingly large amount of memory and complicated programs. In addition, the characterization of details by known methods has a number of additional drawbacks: on the one hand, errors and inaccuracies of the optical registration of the image can cause confusion of the details, that is to say that an error of registration of the image generates details Apparent that they do not really exist, and on the other hand the existing details can not be recognized due to an inadequate registration of the image. Additionally, the current image of a person's finger lines may contain detail errors, for example due to a skin lesion, pollution or a poor ability to register the skin lines, so that the image appears, for example, interruptions. For example, due to a simple injury in the form of a cut along the edge of the cut, a large set of different details can be formed, specifically apparent ends of crests. Therefore, the recorded image of a person is not always identical, which again requires complicated evaluation programs. For these reasons, the determination of fingerprints by known methods of detail requires a lot of effort in terms of calculation and memory. Other known methods for determining identification keys based on the distances between the crests or on the basis of gradients have so far not been able to provide sufficient recognition security with short keys. On the other hand there is a considerable demand to identify and verify people with simple means to be used in a large number of applications of daily use, for example selective access for payment by credit cards, for identification for legal or social purposes , for example the control of passports or for the inspection of personal documents, for example for social programs, etc. For all these non-forensic applications it would be necessary to find a simpler and more secure biometric identification key that requires very little memory and therefore can also be used for economic data carriers. Particularly for economical magnetic cards, for documents with uni or two-dimensional bar codes, or in other economic data carriers, in particular also for chips with EEPROM (read only, erasable and electrically programmable by the user) memory of smart cards and data support systems without contacts. This is also absolutely necessary for all applications related to the handling of businesses involving relatively small sums of money, for example in the field of everyday consumer goods, with vending machines with a relatively large number of users. For this type of applications the identification support must be very economical, that is, it must be able to be applied safely with little memory capacity and a relatively simple evaluation in the small local computers of the verification stations. Accordingly, the object of the present invention is to create a method with a better proportion of the amount of computation key and effort required in relation to the precision of the determination, and in particular that of generating a shorter and simpler key at a sufficiently high determination precision, the key may comprise less than 100 binary digits, for example only 36 binary digits or even less. The key should also be less sensitive in terms of image errors and registration errors as well as in the choice of the section of the image. In addition, the generation of this key should also be possible at local stations with simple inexpensive computers. According to the invention the object is achieved by a method according to claim 1 and a device according to claim 25. By using at least two of the independent or orthogonal features that are distance L of ridges, gradient G , curvature C and bifurcation B is achieved substantially a multiplication of the precision of the determination of the two characteristics, and with the determination of the compressed characteristic values of the frequency distributions of the characteristics are determined the characteristics of the key in a more simple, which is also less affected by registration and image errors. With the method according to the invention it is also possible to achieve a greater precision of the determination based on few characteristic values or very short identification keys for simple applications, or respectively they can be achieved if required by increasing the number of features or of the length of the key. Further convenient developments of the invention are indicated in the dependent claims. The invention is described in more detail in the following description of the various steps of the method as well as in connection with the figures and examples, with figure 1 showing a determination of the distances L of ridges in the x direction, the figure 2 shows a determination of ridge distances from grayscale value images, Fig. 3 shows a determination of L in the x and y directions, Fig. 4 shows frequency distributions HL as a function of length L of the distance, figure 5 shows a classification with determination of class values, figure 6 shows a determination of gradients G, figure 7 shows a determination of curvatures K, figure 8 shows a determination of class values of the distributions HG of gradients of different images, figures 9 to 11 show different images of fingerprints, figure 12 shows a determination n of bifurcations B in a skeletonized image, figure 13 shows a determination of bifurcation distances LB figure 14 shows a determination of bifurcation areas F, figure 15 shows another example of a determination of bifurcations B, figure 16 shows a segmentation and coverage of an image with a grid, figure 17 shows representations of possible sections of images, figure 18 shows a diagrammatic representation of the inventive method, figures 19a, b show examples of defined bifurcations, figure 20 shows a device according to the invention for carrying out the method, figure 21 shows an illustration referring to the classification of the bifurcations. For the determination of the image, a digital image of gray scale values of a fingerprint with a suitable grid of, for example, 512 x 512 pixels (figures 9 a) is recorded by known methods. eleven) . This digital image can be used either directly for the determination of the characteristics or from it an image of the fingerprints can be created by reprocessing the image, in particular by binarization and skeletonization (figures 12 and 15). Then, different characteristics of this image are collected once created, and compressed to an identification key C in additional processing steps, a key that corresponds to the desired application in terms of its length and determination precision. Characteristics of the fingerprints The following four characteristics are used substantially independent or orthogonal respectively, and their frequency distributions are determined: L ridge distances G gradients K curvatures B bifurcations Distance L characteristic of the ridges The distances L of the ridges (lengths distance) are defined, as illustrated in FIGS. 1 and 2, as distances between two successive finger lines (crests) 5, where the finger line has a width 0, that is, corresponds to the distances between the centers of two lines successive fingerprints that are recorded in the direction of a projection X-ray. For skeletonized preprocessed images, this corresponds to a peak distance (with peak width 0, Figure 1), while with images of gray scale values according to Figure 2 between two successive finger lines, the distance L should be calculated as follows: L = L2 + 1/2 x (Ll + L3). If Ll and L3 correspond to the widths of the finger lines with a threshold value of 10 of the gray scale properly selected. With this threshold value of 10, a binarization can also be carried out by designating the gray scale values De greater than the threshold value by 1 and the gray scale values lower than the threshold value by 0. For the elimination of errors of the individual picture elements can be predetermined as a condition that the widths Ll, L3 of the crests and also the distance L2 must amount to three successive picture elements. The determination of distances Lxl, Lx2, etc. which occur in the direction of a recording beam, for example the abcisa x, is carried out as shown in figure 1. The complete image is then recorded by varying and in appropriate stages d and according to figure 3, so that the amount of all recorded distances HLx (x, y) can be plotted as a function of distance lengths Lx according to figure 4. This shows the frequency distribution or histogram of all distances Lx occurring in the x direction over the registered image area. In a way analogous to this, the distances Ly (x, y) in the orthogonal direction (that is, in the direction of the ordinate y) are determined and recorded over the entire area of the image by varying x with the distances dx selected. This leads to an HLy histogram, again over the entire area of the image. To determine the distance lengths Lx and Ly in the x direction and in the direction y, the grid must be oriented in a defined way: in this case with the y axis so that it corresponds to the longitudinal axis of the finger in order to obtain defined histograms . The histograms HLx and HLy are completely different, as can be seen in figure 4. Figure 5 shows a classification of a HLx histogram, being that for class He = 1,2,3 ..., for example the average values Hq. , the maximum values Hmax and standard deviations or variation Hvar are determined and used as characteristic values Ci of the identification key C. The histograms HLx and HLy can be classified each, for example, in 8, 12 or 16 classes, and then one, two or three values (Hq, Hmax, Hvar) can be determined for each class. Characteristic gradients G In analogy to characteristic distance lengths, the histogram of the gradients G, that is to say of the first derivations of the direction, is also recorded regularly over the entire region of the image, and for this purpose it is recorded again, for example, according to the direction of the projection. The gradients are determined as tangents to the line 5 of the skin at the point of intersection of the projection ray (for example, in the x direction). As shown in Figure 6, the gradients are recorded by the direction of the projection, and his histogram HGx is recorded for the projection direction x over the entire region of the image, and in analogy to this the histogram HGy of the gradients in the direction and the projection in order to register regularly the G (x, y) directions of the gradients over the entire region of the image. This histogram of G can also be recorded in a form that covers the image by determining a gradient value for each grid of a grid (30, figure 16) in order to register a grid of gradients. This is explained with more detail later. As an illustrative example, the HG histograms of the gradients were determined for the three fingerprint images of Figures 9, 10 and 11, and are shown in Figure 8. Image II of Figure 9 shows the fingerprint of a person 1 , and the images I2a and I2b show the fingerprints of a second person, wherein respectively Figure 10 shows the fingerprint I2a without errors and Figure 11 shows the same finger of the same person in the image I2b with errors or injuries. The HG histograms of these images were recorded over the entire image and are shown as a function of the gradient angle from 0 to 180 °. Based on this, a classification was carried out in 16 classes, that is, each class with 180 °: 16 = 11.25 ° of angular region, and the average values Hq of each class were determined and are shown in figure 8. result is a graph with 16 characteristic values for each image. As can be clearly seen, the graphs of these images I2a and I2b according to figures 9 and 10 are almost identical; specifically, with a correspondingly defined threshold value S, both images are classified as identical. Accordingly, the identification of this person 2 is still possible even when the images I2a and I2b differ, firstly due to errors 20 due to injury and secondly due to image regions not identically recorded. As shown in Figure 11, the region of the registered image I2a does not correspond at its edges to the region of the image I2b (differentiated definition of the region of the image). In the case of an evaluation of a known detail many new pseudo-details (ridges ends) would occur in the image I2b in the injured areas 20, and therefore the identification would be added to a lot of effort or even impossible, respectively. The graph of the image II of the person 1 of figure 9 visibly differs considerably from the graphs I2a and I2b of the person 2. Accordingly, the 16 characteristic values Ci provide a relatively good contribution to the security of recognition of the key C of identification as partial key Cl of characteristic G. This example also illustrates that the inventive determination of the histograms of the aforementioned characteristics over a wide region of the image, and from there the determination of compressed characteristic values results in an identification key which is only affected to a relatively small degree by the local errors of the image, so that a relative increase in the security of the recognition is achieved accordingly. This is in contrast to the evaluations of known details. As previously explained in connection with the determination of the distances L, the characteristic G gradients can also be determined directly from digital images of gray scale values, for example by determining a gradient value for each mesh of a grid 30, and consequently by a gradient grid that is explained in connection with Figure 16. Characteristic curves K The curvatures K according to Figure 7 are determined at the points of intersection of the lines of the skin or lines 5 fingerprints with projection directions x and y as second derivatives of the direction of the finger lines. This is determined, for example, as the reverse radius R of the approach cycle to the finger line 5 at the relevant intersection point. Analogous to the determination of the characteristics described so far, in this case the histograms of the curvatures are again determined over the entire region of the image in the two orthogonal directions x and y, specifically HKx and HKy (or by determining the K value for each 30 mesh). In order to exclude very small irrelevant radii of curvature that can be formed due to the irregularities in the selection of a skeletonized image, rules can be applied, for example Rmin = 0.3 - 0.5 mm, so that the radius of curvature is included narrower in the center of the image, but nevertheless not even narrower curvatures, for example of the bifurcation B5. As previously explained in connection with the determination of the distances L, the characteristic G gradients, and possibly also the K curves can also be derived from digital images of gray scale values, or from binarized images not yet skeletonized. In accordance with the invention, only the independent characteristics are used, which are the distances L of the crests, the gradients G and the bifurcations B (bifurcations of the lines of the skin), where at least two of these characteristics have to be used for the determination of the key. The characteristics L, G, K are not characteristic details; They are characterized by their histograms. The special characteristic of the bifurcations B is also chosen and includes in a way that the image errors, for example the absence of a single branch, for example the margin, - which can be registered once and not the next time - do not have a substantial effect. It is important that the short identification key C that is wanted is so independent of individual errors of the image or of a singular characteristic (for example of different individual details) as possible. For this reason the only characteristic detail that is used as a characteristic can only be the B-branches clearly defined, being that in contrast to other details these bifurcations B are relatively insensitive to the errors of image or to apparent errors due to disturbances of the finger lines, for example due to cut injuries. Cutting injuries can create new ridge ends as apparent details, but nevertheless they can not (create) bifurcations. It is important that the bifurcations B clearly defined characteristics that are applied here are used in a different way than in the conventional evaluation of the details. The conventional evaluation registers different types of details, and of each detail its position, its type and its orientation, being that it resorts to the relative positions of these different individual details for the evaluation and the determination of the key. However, according to the invention only one type of detail is selected, clearly defined bifurcations, which are additionally used in a totally different manner than hitherto. This is explained in the following: as histograms or as bifurcations individually skeletonized. Bifurcations B characteristics Figure 12 shows a binarized or skeletonized representation of the grayscale values image of Figure 10 which is used as an example for the determination of bifurcations. In this image the bifurcations Bi = Bl - B12 are determined, which results in an amount of N = 12 bifurcations B. For this purpose, selection rules or respectively adequate definition criteria are established, so that the bifurcations are not counted as bifurcations. small errors and other types of details (for example, islands). The rules are, for example: a registered branch must have a minimum length of 0.5 - 1 mm of all three branches, and a branch must have a length of at least 1 - 1.5 mm. In addition, a minimum distance between two branches of, for example, 0.7 - 1 mm can be prescribed. According to such definition criteria, they are not counted as bifurcations, for example, (B13), (B14) of Figure 12 and (B7), (B8), (B9) of Figure 15. Branches B can be used as individual or selected defined bifurcations to generate a key, as explained in connection with Figures 19 and 21 or, in accordance Figures 13, 14 can be used to determine histograms, for example of branch distances as well as adjacent triangular areas F between the branches. According to figure 13 the distances LB of the branches are determined as follows: The distances LBi-j between the N branches B and each of the other branches Bj are determined.

This results in N (N-l) bifurcation distances LBi-j. In the example with N = 12 this results in 132 distances that are plotted in a histogram: as HLB distribution of the frequency as a function of the distance LB. In analogy to this, the triangular areas Fi-j between the branches Bi and Bj are determined according to figure 14. A first triangle side is defined that extends from each bifurcation Bi to each other branch Bj, the third point being of the triangle is determined to be the Bk fork (with k no = j) closest to the bifurcation Bi; for example, starting from Bl, the areas Fl-2-3 (with i = 1, j = 2, k = 3) and Fl-3-2 (with i = 1, j = 3, k = 2), as well as Fl-4-2 to Fl-12-2. With this selection rule, it is guaranteed that only one area is registered per bifurcation, for example, in the case that the two nearest branches are Bk branches, only the one closest to Bj is used as Bk. Starting from BIO to Bl the area F10-1-12 (i = 10, j = 1, k = 12), that is, starting from each Bj as a baseline, (this) results in precisely one triangle, therefore again a sum of N (Nl) = 132 Fi-j areas, which again they form a histogram with the distribution of the frequency as a function of the area F.

Figure 15 shows another example for the determination of the bifurcations of another fingerprint image, being that by means of the corresponding selection rules, for example that bifurcations B7, B8, B9 located very close to one another are not counted as branches for the evaluation. on the other hand, so that in this case this leaves only the branches Bl to B6 and consequently the quantity of the distances LB of the branches as well as that of the nearest triangular F areas is each N (Nl) = 30. With other definition criteria, for example, only B9 could be defined as bifurcation, but not B7 and B8. Determination of the characteristic values From the histograms, simple Ci characteristic values characterizing the histogram are determined, for example average values Hq, maximum values Hmax and variations Hvar. In addition, a histogram can also be classified into He classes and the average value and variation, for example, can be determined as characteristic values (Figure 5). Characteristics of selected bifurcations The characteristic bifurcations can be used additionally in a non-statistical way by the use of clearly defined individual bifurcations with certain bifurcation characteristics for the determination of the characteristics Ci of the key and the identification key C. In this the bifurcations are determined by their positions, as illustrated in the examples of FIGS. 19a and 19b, that is, the local coordinates xl, yl of the point Pl are determined as well as the angle of orientation a and the angle Wb of opening. A clearly defined opening angle Wb can be defined, for example, as the angle between the connecting line of the points Pl, P2 and Pl, P3, where the points P2 and P3 are chosen at a suitable distance rl from the point Pl , for example rl = 0.5 - 1 mm. A point P4 with a minimum distance r2 of, for example, l to 1.5 mm from point Pl, on the other hand, forms a definition criterion for a branch, so that small errors are not counted as bifurcations. In the example of Fig. 19a a relatively small bifurcation angle Wb is shown, which corresponds, for example, to the bifurcation B6 of Fig. 2, as long as the example of Fig. 19b shows a large aperient angle Wb, for example. example similar to that of branch B8 of figure 21. Now the bifurcations identified can be classified for the determination of the key, that is to say that classification criteria are introduced, as for example a valence for the bifurcations. A first classification criterion is the opening angle Wb, where the larger opening angles have higher valences. The criteria for classification can be, for example: a valence in proportion to the opening angle Wb. Another additional classification criterion is the distance RBi of the bifurcation towards a central point of reference, in this case, for example, the distance RBi to the center Km of the curvature (figure 21), being that the central bifurcations have greater valences than those more distant. As an additional selection criterion, bifurcations that are close to the margin, for example B2 in Figure 21, which are close to the external contour A of the image of the finger lines can be omitted. From the characteristics of the branches and the classification criteria you can define, for example, a branch priority as BP = Wb / RB. With this, all bifurcations can be classified in descending order according to BP. To generate a short key length, the first entries of the classified list can be used according to the priority of the branch. A possible order for Figure 21 can be, for example, the following: B8, B6, B5, B7, B4. Through the combination of characteristic gradients and bifurcations, particularly concentrated keys can be created with relatively high recognition security. An identification key C is particularly convenient, which on the one hand consists of gradient characteristics for each IS segment of a gradient grid, and on the other hand of the bifurcations B classified. For this purpose, a central point of reference and the orientation of the image of the fingerprints are required. Preprocessing the image The L and G characteristics can be determined directly from the grayscale image image without additional pre-processing of the image. The determination of the characteristics B and also of K most of the time is carried out by means of a thinning of the lines as preprocessing of the image, for example binarization and skeletonization. Then the other characteristics L, G, K are determined from the skeletonized image. However, in a convenient variant the bifurcations B can also be determined directly from the image of gray scale values by means of neural gratings. The characteristics can be recorded in the following ways, as illustrated in figure 16, with segmentation and classification of the mesh: a in the entire region of the image (also for the characteristics L, G, K, B) b in each one of a few relatively large segments, for example divided into 3 x 7 to 5 x 7 IS segments, whereby histograms are determined for each IS segment and from them characteristic values c are derived by covering an I image with a grid 30, that only one value of the characteristic is determined for each mesh. The size of the mesh is selected appropriately, (for example 5 x 5 pixels). With this type of register a, b, c the following characteristics are recorded: a b c Characteristics L - - G G G K K K B - (B) When comparing the keys, attention should be paid to the fact that the division into IS segments depends on the translation and rotation of the recorded images. This must be taken into account for the comparison of the keys.

The characteristic K curves are independent of the rotation, the same is valid for the lengths BL of the branches and the areas F of the branches derived from the bifurcations B and the relative position of the selected branches. The characteristic G gradients require a determination of the axis and orientation of the image. On the other hand, the characteristic G gradients are independent of the image expansions (amplification or reduction of the image). By virtue of the fact that in the identification key C = (Cl, C2) are contained at least two characteristics (L, G, K, B) as partial keys (Cl, C2), these partial keys can also be used in a manner correspondingly different for the evaluation of the key, for example in a manner corresponding to the dependence of the image definition and to the dependencies of the respective characteristic and the evaluation of the selected histogram or bifurcation characteristics, respectively. Partial wedges may be formed at the discretion of any characteristic values Ci. Section of the image Figure 17 illustrates the dependence of the definition of the image, being that two images of the same finger were recorded, first an image 13 and then an image 14, these images 13 and 14 show different regions and possibly also Different orientations of the axis and longitudinal fingerprint. In order to always be in a position to compare similar image regions it is also possible to cut out a central region of a specific fingerprint image, which portion can then be used as the IA section of the image to determine the characteristics and generate a key . In this, the regions of the center Z of a fingerprint and its surroundings ZU are conveniently used for the recording of the characteristics and the generation of a key, in virtue of which the information content is higher there than in the marginal regions. With a well-defined position and orientation of the image by corresponding image recording devices, for example a two-finger guide, also the segmented images can be compared with greater security. Oreintation and central point of reference For determination of the orientation and of a central reference point of a fingerprint image, or for the definition of a grid with a origin of the coordinates, it can be defined in a manner known per se, for example , what is known as a nuclear point. However, this goes hand in hand with great effort, and often a clear nuclear point can not be determined. A simpler method is to determine a center of gravity Km of the radius of curvature as the central point of reference from the characteristic curvature. For this purpose the centers of the approach circles are determined (see R in Figure 7) to the curvatures, and from these regularly recorded circle centers of the approach circles the centers of gravity Km are determined with the coordinates xm, ym . However, a central reference point can also be determined first by approximation, for example as the center of gravity of the area of the image with the outer contour A (in Figure 21). Similarly, the registration of a fingerprint recorded in a defined manner by guiding elements 32 can be used as described in connection with FIG. 20. Another possibility is to determine a central reference point by the Hvar variation of the gradient distribution in FIG. the IS segments of the image, being that the central point of reference is determined by the segment that has a maximum of variation. For this purpose it is also possible to vary the segmentation of the image. With an inappropriately defined position of an image to be registered it is necessary to select the characteristics, ways of determining and determining the key so that they are relatively independent of the position, for example the characteristics G, K according to the form of determination, although in most cases the address should be determined as support for. For the comparison of the key the image can also be rotated around a small angular region or moved in the x and y direction. As variants of implementation of the inventive method, depending on the image definition and the objective pursued, the following combinations of characteristics can be used, for example, to determine an identification key C: 1. The characteristics that are distances L of the crests, the gradients G and the bifurcations B. These can also be, for example, determined from images of gray scale values and used in particular for keys (L and G) that are shorter and simpler, for example also with a classification of histograms. From the skeletonized images the following characteristics can be used: 2. The characteristics that are the gradients G and the curvatures K 3. The characteristics that are the gradients G and the bifurcations B 4. The characteristics that are the curvatures K and the bifurcations B that are both invariable to the rotation 5. The characteristics that are the gradients G, the curvatures K and the bifurcations B, this in particular for a greater precision, being that correspondingly to the greater precision longer keys are formed (also with selected bifurcations). After a suitable choice of the combined characteristics, it is possible to choose the precision with which the characteristics and their histograms are determined, as well as the determination of the characteristic values Ci from this, so that finally an identification key C is formed with a key length with the required precision demands, for example it is possible to lengthen the length of the key as required by a narrower segmentation and classification and by determining more characteristic values Ci. The identification key C can be determined not only sequentially but also iteratively. Key Comparison With these relatively compact inventive identification keys C (which require less memory), all identification, verification and authentication tasks can be carried out in a local, simple and rational manner. In this for the comparison of two keys Ca and Cb, for example a key of a recent fingerprint image Ca with a reference key Cb of an associated memory (which can be a data bank or also an identification means IM) a key difference D = Ca - Cb is determined by mathematical methods and compared with a predeterminable threshold value (acceptance value) S. In the case where D <; S it is verified that the two keys Ca and Cb and consequently the corresponding persons are identical. Through this, individual threshold values specific to people can be given. For example, depending on the precision with which you want to determine the identification key C of a certain person and, on the other hand, also depending on the importance of the identification task for this person. For the determination of this correspondence, for example, the Euclidean distance D of the characteristic C vectors with a predetermined correspondence threshold S can be determined. This means that for a key formed by the characteristic values Ci a fingerprint 1 is identical to a fingerprint 2 when: D = S (Cli-C2i) 2 < S Alternatively the identification keys can also be correlated according to the following method: A Kor correlation is formed between the characteristics of a finger ai with i = 1, N and a second finger bi with i = 1, N according to the formula: -l = Kor = S (ai-bq) (bi-bq) / (S (ai-aq) 2) 1/2 < + l for i = l, N being that aq is the average value of the values ai and bq is the average value of the bi values. For ai = bi the fingers are identical, that is to say that the correlation is 1. In Kor = 0 the characteristics are not correlated, in Kor = - 1 they are counter correlated. Other possible methods of comparison are the regressive analysis or the momentary analysis depending on the type of the key C. In addition, the different characteristics or characteristic values Ci and the corresponding partial keys Cl, C2 of a key C = can be treated differently. (Cl, C2) of identification. Figure 18 illustrates diagrammatically the inventive method for determining a relatively short identification key C. The choice 52 of at least two of the orthogonal characteristics L, G, K, B as well as the choice of one of the following steps of the method on the one hand is based on the definition 51 of the image and on the other hand on the task 60, that is to say in the required recognition security of the desired applications, and thus in the erroneous acceptance rate FAR required and the erroneous rejection rate FRR required. Depending on the definition 51 of the image (ie the quality of the orientation, section, image registration and image quality), characteristics and types of registration correspondingly less sensitive are chosen. It is also possible to carry out a preprocessing 53 of the image for a direct determination of the characteristics from a grayscale image, a binarized image or an image of skeletal rarified lines. In a following step 54, an orientation of the image, a central reference point or an origin of the coordinates, for example Km, as well as respectively the choice of a section IA or an IS segmentation of the image may follow. From this, H histograms are respectively determined with a fine resolution or corresponding data quantity. In addition to the histograms, bifurcations B individually defined characteristics with position, angles Wa of orientation and opening angles Wb can be defined in step 56 and in step 57 a classification of the bifurcations, a skeletonization of the valences can be carried out respectively. according to a bifurcation BP priority. In step 58, the determination of the characteristic values Ci is carried out and in step 59 the composition of the identification key C is carried out. In other words, a correspondingly longer or shorter key length is made up until the desired FAR and FRR recognition security is achieved according to step 60. The methods of evaluation and evaluation methods can be chosen according to the composition of key C. the threshold values S for the evaluation of the identification. The evaluation of the identification (comparison) Ca - Cb = D <; S is carried out in step 62. Accordingly, the inventive method is universally applicable for a wide range of applications and can also be optimally adapted to the desired tasks in terms of computational effort and key length. Figure 20 shows a device for carrying out the method comprising an electronic image recording device 31 and a station with evaluation electronics 34 and evaluation algorithms 35 for the determination of characteristic values Ci and from these the key C or Ca of identification respectively, ie of the current identification key corresponding to the registered image of the fingers 1, 2 to be recorded and to carry out a comparison of the keys Ca-Cb between a stored reference key Cb and the current Ca key. This comparison of keys can also be carried out in a corresponding WR reading station. Corresponding to the comparison of the key, that is, to the current verification of the person, access functions can also be practiced for the control of corresponding operating stations. The image recording device 31 in this case comprises guide and positioning elements 32 with limit stops for orientation and positioning of one or two adjacent fingers 1, 2 to be recorded. Additionally, a lateral limit stop can be provided to position the longitudinal axis of the finger and a front limit stop for the placement of the finger image and therefore for the determination of a central reference point or a center of the finger image respectively. This type of two-finger recording device is disclosed, for example, in PCT patent CH 97/00241 = WO 98/09246. With a device according to the invention, a comparison of Ca-Cb keys can be carried out locally so that a central data bank with the reference keys Cb is not required. The device according to the invention can also comprise corresponding identification means IM for authorized persons, being that in the identification means IM or in its memory 43 the reference key Cb is contained and additional information, by means of which it can be taken to performed a communication 40 encoded with a corresponding WR reading station. In this the reference key Cb of the authorized person is only stored in the identification means and not in the verification station or in a data bank respectively. This allows a better protection of the data. In a further variant it is also possible to carry out a key comparison Ca-Cb, ie of the current key Ca against the reference key Cb of the person by means of a processor 41 in the identification means itself. In this case it is only necessary to transmit the current key Ca from the reading station WR to the identification means, since it is not necessary to transmit a key from the identification means. In another variant, the device can also be connected to a master system 47 with a main processor and a data bank 48 for the reference keys Cb. The identification means IM may contain access and operation authorizations for the corresponding operating stations 46 of an installation. Pre-selected database search In those applications in which a quick search of the data bank is required for the positive identification of individual persons in a database of inhabitants with a large number of reference Cb keys only by means of biometric recognition features , the identification can be carried out in stages. In this case, the comparison is not carried out with the key Ca of complete biometric identification but only with partial keys or with individual characteristic values Ci, respectively, so that the search can be carried out at a considerably higher speed. In a first stage for example, only a partial Cl key is compared and the remaining reduced quantity is then compared to a second partial key C2 in a second stage, etc. This continues until the complete evaluation of the key with a very small remaining amount of the data bank. This type of search is considerably faster than a search for the full identification C key. Conveniently, the search for primary biométic data is classified, that is to say for example, for this purpose, the gradients and selected bifurcation characteristics are used in stages. Another method is to form segments of gradients according to the method of segmentation of quadrants (quadrant trees). In this the image is divided into four segments from each of which a gradient histogram is formed. In the next stage it is divided again into four quadrants, etc. until finally the gradient histogram converges with the tangent of the main line in the considered segment.

The following terms are used in the description and in the figures: L ridge distance G gradients K bends B bifurcations LB fork distances F fork areas H frequency distributions, histograms H (L), H (G), H (K ), H (B: LB, F) He classification of H Hq average values Hmax maximum values Hvar variations of H Of gray scale value II, 12 fingerprint images IA image sections IS image segments C identification key (characteristic vector Ci) Ca current key Cb reference key Cl, C2 partial keys Ci characteristic values D key difference S threshold value, acceptance value Z center ZU surroundings x abcisa and ordinate, longitudinal axis of the finger dx, d resolution FAR rate of wrong acceptances FRR wrong rejection rate IM means of identification, data support WR reading station BP priority of branch Pl - P4 definition of Wb rl, r2 distances Wa angle of orientation Wb opening angle of B Pl (xl, yl) local coordinates of B Km center of gravity of radius of curvature xm, and m coordinates of Km RBi distance from the center reference point to Bi A external contour of the image finger I 1,2 finger 5 finger lines, skin lines 10 gray scale value 20 image error, 30 grid injury 31 image registration device 32 guide and placement means 34 electronic evaluation, evaluation station 35 evaluation algorithms 40 coded communication 42 processor 43 memory 46 access function (functional stations) 47 master system, main processor 48 data bank 51 image definition 52 choice of characteristic L, G, K, B 53 preprocessing of the image, skeletonization 54 image section, segmentation 55 histogram formation 56 bifurcation determination 57 classification, choice of B 58 determination of the characteristic values Ci 59 determination of the key C of identification applications, demands (FAR, FRR) method of evaluation, choice of threshold values comparison of identification Ca - Cb

Claims (25)

1. CLAIMS 1. Method for determining an identification key from fingerprint images or grayscale image images characterized by recording at least two of the following independent characteristics: crest distances, gradients, bifurcations , and that from at least one of the characteristics are determined H histograms from which of which are determined characteristic values such as the average value, the variation and the maximum value as compressed characteristic values that as vector components form the key of identification. Method according to claim 1, characterized in that the characteristic gradients are used. Method according to claim 1 or 2, characterized in that the histograms are determined in at least two orthogonal directions x and y. Method according to one of the preceding claims, characterized in that the image is preprocessed, for example by means of binarization or skeletonization. Method according to one of the preceding claims, characterized in that a classification of a histogram is carried out, since at least one characteristic value for each class is determined as a characteristic value. Method according to one of the preceding claims, characterized in that the registered image is subdivided into several segments and that the histograms of the characteristics of length, gradient, curvature and bifurcation are determined for each segment. Method according to one of the preceding claims, characterized in that the identification key is derived from a section of the image comprising the center of the image of the fingerprint and its surroundings. Method according to one of the preceding claims, characterized in that the length of the identification key is selected in such a way that the accuracy or security requirements of the recognition are met respectively for a desired application. Method according to one of the preceding claims, characterized in that the partial keys of the characteristics are determined differently and / or used differently for the evaluation of the key. Method according to one of the preceding claims, characterized in that individual bifurcations defined as characteristics are used. Method according to claim 10, characterized in that the characteristic branches are determined directly from the gray scale digital image by means of neural gratings. Method according to claim 10 characterized in that the characteristic branches are classified according to their opening angle and / or according to their position in the fingerprint. 13. Method according to claim 10, characterized in that a bifurcation priority is defined for the classification, the bifurcation priority being increased as the opening angle increases and the distance to the center of the image decreases. Method according to claim 10 characterized in that the branch priority is defined as BP = Wb / RBi. Method according to one of the preceding claims, characterized in that the reference center point is defined as the origin of the coordinates of the fingerprint image. Method according to claim 15, characterized in that a center of gravity of a radius of curvature is defined as a central point of reference. Method according to claim 10 characterized in that the variation of the gradient distribution in the image segments is determined and the reference center point is determined by the segment having a maximum variation. Method of compliance with one of the preceding claims characterized in that when the image is recorded, at least approximately the longitudinal axis of the finger and a central reference point are determined by means of guidance and positioning. Method according to one of the preceding claims, characterized in that the characteristics are used, which are the gradients and the bifurcations. Method according to claim 19, characterized in that the identification key is formed from the characteristics of the gradients for each segment of a grid of gradients and classified branches. Method according to one of the preceding claims, characterized in that partial keys are formed and that a comparison Ca-Cb is carried out with a large number of reference keys of a data bank in stages corresponding to the partial keys. 22. Method according to claim 21, characterized in that the step evaluation is carried out by means of gradient segments and the subdivision method of quadrants. Method according to one of the preceding claims, characterized in that a current identification key is registered and determined and compared locally by a corresponding reading station with a personal reference key stored in a corresponding identification means. Method according to one of the preceding claims, characterized in that the identification key is determined sequentially and / or iteratively. 25. Device for carrying out the method according to one of the preceding claims, characterized in that it comprises an electronic image recording device., a station with evaluation electronics and evaluation algorithms for the determination of an identification key and for carrying out a key comparison with a stored reference key, as well as with an access function for the control of the corresponding functional stations . Device according to claim 25, characterized by guide and positioning means with a lateral limit stop and a front limit stop for orienting and placing one or two adjacent fingers to be registered. Device according to claim 25 or 26, characterized in that the comparison of the keys is carried out locally. Device according to one of claims 25 to 27, characterized by a corresponding identification means of an authorized person containing the reference key of the person and additional information for the communication in code with a corresponding reading station. Device according to claim 28 characterized in that the current key is transmitted to the identification means and because the comparison of keys with the reference key is carried out by the processor of the identification means. Device according to claim 25, characterized by a master system comprising a main processor with a data bank for the reference keys and local operating stations. Means of identification for an authorized person with a reference key of the stored person, key is determined with a method according to one of claims 1 to 24, for the communication in code with a corresponding reading station. Identification means according to claim 31 with access and operation authorizations for the corresponding operating stations.