FI103013B - Biometric identification procedure - Google Patents

Description

103013

Biometric identification method

The present invention relates to a biometric identification method, wherein one or more fingers of a human palm are provided with an identification-5 information and the identification information is compared with one or more known reference models, and if the identification information corresponds to the reference country, the identification is performed.

Biometric authentication methods are used, for example, in Access Control 10 to identify people who have access to a controlled space, building, or other similar controlled object. In biometric methods, the target may be, for example, the retina of the eye, sound or palm.

In known biometric identification methods, where the subject is a human palm, the identification data and reference patterns are measured from fingerprint, palm geometry, knuckle location, or blood vessels. In the known solutions, only fingerprints have been measured on the fingers, i.e. the measurement has been made on the inside surface of the palm. On the back of the palm, blood vessels or knuckles, ie no fingers, have been measured.

US-5,082,001 discloses a device for measuring parameters particularly related to upper limbs or scars. The gauge comprises a computer and a stick, which the therapist can use to determine, for example, the path of movement of the hands or monitor the scar on the skin.

WO-88/04153 discloses a method and apparatus for identifying an individual by forming an image of a selected portion of an individual's body and extracting information such as a subcutaneous vascular pattern and comparing said information with previously defined information.

US 4,206,441 discloses a device for identifying an individual based on palm pattern measurement. The device first generates an ana-30 logical signal from a palm image, for example, with a TV camera, and then said image is converted into a bit pattern from the palm outline. The resulting contour information is processed such that the convex portion of the palm curve code information is used for personal identification.

US 4,186,378 discloses a method for identifying a person and a pattern recognition device. In the system, a person's palm is imaged with 2,103,313 cameras (e.g., vidicon or CCD), the resulting signal is ditferenced to highlight peaks and valleys, and then the signal is quantized and stored. The generated data is analyzed, and the data generated from the positions of the representative patterns is sorted and stored in its own table.

DE-2 720 151 discloses a device for reading the shape of a palm.

WO-94/22370 discloses a method and apparatus for identifying an individual from an image of subcutaneous vascular patterns.

US-A-5,073,949 discloses a device for identifying an individual based on a fingerprint image. The image is formed especially with the fingertips.

US 4,468,807 discloses a method for analyzing the details of a stored image. The method specifically deals with the change in the details of the palm 15 due to the location, age, disease, temperature, etc. of the palm.

U.S. Pat. No. 4,614,366 discloses nail attachment identification! iuska. The method involves attaching a bar code strip to the nail, which is read and used to identify the individual. The barcode is industrially manufactured and therefore easily counterfeit, unlike the human's natural identifier.

According to the Applicant, known methods have problems with the speed, accuracy and reliability of the methods, both in the implementation of the hardware and software solutions required and in their cost. The problems are due to the measurement point and the reference algorithms.

It is an object of the present invention to provide a new type of method which avoids the problems associated with known solutions.

This object is achieved by a method according to the invention, characterized in that the identification information is formed on the back of the palm by one or more of the following: sor-30 mini-wrinkles, wrinkles between finger joints, nail stroke.

The method of the invention is based on the idea of utilizing in the palpation the wrinkles on the back of the palm, the wrinkles between the finger joints or and the dash on the nail, all of which carry a humanly identifiable identifier consisting essentially of parallel wrinkles or ribs.

3, 103013

The process according to the invention achieves several advantages. In the method, both identification data and reference models can be easily and simply generated. The method is fast in terms of measurement technology and computation and also gives very reliable measurement results. According to the Applicant's observations, the probability of errors 5 is less than 5/106.

The invention will now be explained in more detail with reference to the accompanying drawings, in which Fig. 1 shows a measuring arrangement used in the method, Fig. 2 shows a back view of a palm, Fig. 3 shows a partial view of a human finger 20 different identification lines of the identification image, Figure 5 shows a grayscale diagram of the 1-line reference model or comparator 15 shown in Figure 3, Figure 6 shows a grayscale 20 Figure 8 shows a gray scale diagram of the difference between the identification image line 10 and the reference pattern image, Figure 9 shows a piecewise correlation.

The measuring arrangement according to Figure 1 comprises a measuring device such as a camera device 1, a data processing unit 2, a lighting device 3, a diffused light shield 4 • · · 25 and a hand hub 5 supporting the palm 6. The camera device is most preferably a matrix camera. The data processing unit 2 may be, for example, a microcomputer comprising an image processing card and analysis software. In principle, the measuring device could also be, for example, an optical rangefinder, but according to the applicant, the camera device 30 is a better choice because the camera device does not need to be moved. The application is represented by reference numeral 7 and is, for example, a door lock device in the case of an access control application that monitors only the desired persons from a locked door.

Figure 2 shows a view of the back of palm 6, in figure 2 palm 35 6 comprises fingers 11-15. Fingers, such as the middle finger 13, comprise two finger joint hinges 16a and 17b in the two finger hinges 13a and 13b, and the hinged grooves 16a appear darker and the lighter bulges 16b appear. The finger articulation ribs 17, respectively, comprise the dark articulated articulation grooves 17a and the lighter articulated projections 17b. In addition, the finger comprises wrinkles 18 between the seam joints 13a, 5 13b, i.e. also between the finger joint sets 16 and 17, comprising darkened grooves 18a and lighter visible projections 18b. Further, the finger 13 comprises a nail 19 comprising a line 19a comprising lines 19b.

As with biometric authentication methods in general, in this method 10, one or more fingers such as finger 13 of the human palm 6 are generated identification information and the identification information is compared to one or more known reference models and the identification is performed if the identification information corresponds to the reference country.

According to the invention, the identification information is formed from the finger 15 on the back of the palm in the region of one or more of the following points: finger joint 16 and / or 17, wrinkles 18 between finger joints 13a, 13b, nail line 19a. The reference models are formed at the respective points. It is common to the creases 16, 17 and 18 and the nail line 19a that they all comprise parallel substantially parallel deflections, such as grooves 16a, 17a, 18a, 20 protrusions 16b, 17b, 18b and lines 19b which can be utilized in identification according to the invention.

In the following, particular reference will be made in the examples to preferred embodiments in which the identification information, or identification pattern, and reference patterns, or reference pictures, are measured, for example, from the creases 16, or grooves 16a, and According to the Applicant's observations, the best result is the measurement of the wrinkles 16 and / or 17 at the finger joint, such as the joint 13a and / or 13b, because they are the most distinguished.

Fig. 3 is a longitudinal partial view of the human finger 13 from the back of the palm 6 and a 1-line reference line 30 of the reference model, which forms a so-called. reference pattern 30, i.e. most preferably reference image 30. Reference image 30 comprises, for example, 640 pixels or pixels. Such comparisons 30, most preferably 1-line, are stored in a large number of computer 2's memory as a database. For example, a reference type of Figure 3 is stored in memory, that is, most preferably, a 1 line image line of all persons having access right past the val-35 control point. Fig. 5 shows a gray scale chart 30 ', i.e. a recognition signal, of a 1-line reference model 30 5 103013 in Fig. 3. A high grayscale value is a light spot and a lower grayscale value a darker spot, so that in Figure 5, the peaks of the signal 30 'are generated by the creases 16b of the wrinkles 16 and the valley points of the signal are generated by the 5 arcs.

Figure 4 shows a partial view of a human finger and an image-limited identification 40 lines of 20 images. The recognition image is also 640 pixels long. The identification image 40 is taken when a person wants to bypass the monitoring point and then has to place the palm 6 on the imaging platform, after which the camera 1 takes a picture 40 from a position according to the invention, for example finger swing 16, which the computer 2 then compares with 30, and if it detects that the identification image 40 sufficiently corresponds to any of the comparator 30, then the recognition is generated and the person is allowed to bypass the selection point 7. Figure 6 shows a 20-line identification information 15

For a better understanding of the reference image 30 and the identification image 40, and the corresponding signals 30 'and 40', Figure 7 is shown showing the grayscale graphs 40 '(line 10) and 30' of line 10 of the identification image 40 and the reference image 30 increased by 50 units. . It should be noted that the grayscale-20 enhancement is made only to illustrate Figure 7 so that the graphs 30 'and 40' do not overlap.

In a preferred embodiment, as can be deduced from the foregoing, the identification information and the reference models are image information consisting of the identification image and the reference model images, since then it is possible to use a camera rather than, for example, a rangefinder. In this case, the identification image 40 comprises a plurality of image lines, for example 20 image lines. Reference models, such as reference model image 30, comprise at least one line of images, most preferably one line of images. The large number of lines in the identification image 40 and the small number of lines in the reference model simplifies and speeds up the process and eliminates problems due to possible twisting or tilting of the finger. The problems result from the fact that if the finger 13 is not in the correct right direction, the identification image 40 does not become exactly parallel to the reference pattern image 30.

The identification image 40, i.e. the detection signal 40 ', shown in Fig. 4 is a so-called "center line" of the finger. A profile consisting of grayscale values of the centerline pixels of the pixels. Such a profile is a signal whose correlation with reference models of data base 103013 can be quickly calculated and used for identification.

Fig. 9 shows a sectional or block correlation according to a preferred embodiment of the invention. In Figure 9, the upper curve 30 'represents a reference reference image (in fact, a reference reference image as a signal) stored in the memory of the computer 5. For the sake of clarity, Figure 9 has the same reference numeral 30 'as Figure 5, although the curve is slightly different in shape. In Fig. 9, the lower curves 40 'represent an identification image (in fact, an identification image as a signal) taken by the camera 1 in the detection situation.

Referring to Figures 3-7, and in particular 9, the Single row of Comparison Figure 30 and the Rows 1-20 of Identification Figure 40 are compared with each other in piecemeal (block) and overlapping fashion. In Fig. 9, the reference picture comprises blocks m1, m2, m3 ..., mn, i.e. windows.

In the method, each block m1, m2, m3 ..., mn, i.e. the window is m kappa-15 flat pixels in length. For example, let m be 80. The number m should be chosen such that the block length is short relative to the length of the row. In the example shown, the ratio of block length to row length is 80/640.

For example, the first block m1 only starts after 20 pixels long initial shift, so that in the method, especially in the first step 20, transitions can also be made backwards. This preferred embodiment enhances the usefulness and accuracy of the method. Thus, since the block length was 80 pixels, the first block m1 thus comprises the reference image pixels 20-100. In the first step of the method, the block or window m1 of the line portion of the reference image 30, 30 'is compared by correlation with the substantially equal lengths of all the image lines, or rows 1-20, of the identification image 40, 40'. Each portion of the image line of block 40, 40 'of the identification image 40, 40' is correlated with block m1. The correlation is performed at the same position as the block m1 of the reference image and also on both sides, for example +/- 10 pixels, around the starting point of the block m1, which was therefore 20 pixels due to the initial shift. For example, if pixels 20-100 of the reference image 30, 30 'are selected as block m1, this window m1 is correlated with all windows in the recognition image at positions 10-90, 11-91 ..... 20-100, ..., 29-109 , 30-110. The result is thus 20 x 21 = 420 correlations. Suppose a correlation calculation would have told that, for example, line 8 of identification image 40, 40 'and line 25-105 on that line gave a par-35 correlation of Hah to block m1. Then, in the second step of the method 7 103013, the starting point is row 8 and the column offset +5. From these 420 correlation results, the highest value, i.e. the maximum, is selected and the starting point of the second step of the method is the line 8 of the identification image corresponding to this maximum and the +5 column offset. The maximum found is stored in the correlation table. The above preferred embodiment of to-5 may also be represented by determining at the initial stage of the piecewise and block correlation which image line 1-20 of the identification image 40, 40 'achieves sufficient (maximum) correspondence to the reference image 30, 30', and the number of lines with which the reference image will be correlated. Thus, at least about the right row and point in that row have been found as a basis for further operations, whereby the number of rows can be reduced, whereby the method is greatly accelerated since the correlation does not have to be calculated so many for subsequent blocks m2, m3, ... mn.

In the second step of the method, the window m1 of the reference window is used for correlation, starting from a fixed and block (15 pixels) offset of a small number (e.g., 20 pixels) pixels relative to the length of block 15 (80 pixels). In practice, this preferred embodiment means that the correlation is also overlapping, whereby in the block correlation the next block of the analogue country is overlapping with the previous block. Thus, block m2 is overlapping, i.e. overlapping with block m1, block m3 is overlapping, i.e. overlapping with block m2, etc. Interleaving overlaps the blocks and ensures that the method is more reliable in producing reliable results. Thus, each block begins at the start of the previous block m1, m2, m3 ..., mn overlapping offset. In a preferred embodiment, the method comprises operating at least 50% of the block length, most preferably about 75%, between successive blocks of the reference model. Thus, the above implies that the blocks of 80 pixels overlap over a distance of 60 pixels if the interleaving transfer is the said 20 pixels. According to the applicant's observations, the overlap of the length shown is sufficient but not excessive.

When proceeding with the method, select the en-. 30 rows resulting from the first step with a small number of rows on each side (for example, 1 row). This reduction of the lines of the identification image provides the aforementioned advantages in terms of acceleration of the correlation calculation. The significance is great because as the number of rows drops from 20 to 3, the need to calculate correlations decreases in the same proportion.

8 103013

The rows selected for correlation in the identification image 40, 40 'are correlated at the same position as where the window m2 is located in the reference image, plus the resulting column transition resulting from the first step. The correlation is made on both sides of the identification image to be correlated at the beginning of the selected rows (position 5 of m2 + column offset +5), but still at a shorter interval compared to the first step, e.g. only + 1-2 pixels. Shortening the correlation interval from +/- 10 to + 1-2 in a preferred embodiment accelerates the method because the number of correlations to be computed is reduced. As with the reduction of rows, so does the correlation interval shortening, since as the 10 correlation intervals drop from 21 to 5, the need to compute correlations decreases in the same proportion.

Also, from the correlation results calculated using the block m2, a maximum is selected and added to the total correlation table. The line corresponding to the maxima and the column offset determine the rows of the identification image to be correlated with the next window, i.e. the m3 block, and their position. The correlation of the m3 window with the identification image is the same as for the m2 window.

For example, if the output of the second step of the method is the results of the first step, row 8 and column offset 5, the correlation is continued so that the window m2 (reference pixels now pixels 40-120) is correlated with rows 7, 8, and 9 of the identification image. The rows selected for the identification image correlation start at 40 + column offset, i.e. now 45. This is how the m2 m2 window is correlated with the identification image positions 43-123, 44-124, 45-125, 46-126, 47-127. The result is thus 3x5 = 15 correlations, which is a significantly smaller number than the 21x20 = 420 correlations of the previous step. Assume that the identification image: 25 line 7 and 43-123 gave the best correlation with the m2 window. Then the result of the next correlation is row = 7, column offset = 5 + -2 = 3. The following correlation is performed in the same way as the correlation with the m2 of the window.

When all n sets of correlated, overlapping blocks m1, m2, m3 ..., mn, i.e. windows, of the reference image are correlated with the selected rows of the identification image, a minimum is searched from the 30 total correlation tables which is the final result of correlation between the reference image and the identification image. The result of the method is a correlation percentage between -100% and + 100%, for example, a detection limit of 70% can be considered.

The rather detailed preferred embodiments described above may be more generally described. Thus, in a preferred embodiment, a block correlation determination (m1, m2, m3, ... mn) is provided, wherein the reference country, i.e., most preferably, the reference image 30, 30 'and the identification information, most preferably the identification image 40, 40'. comparing each other block by block and determining the block partial correlations, from which the final result of the correlation between the reference model 5 30, 30 'and the identification information 40, 40' is determined. In a still more preferred embodiment, the method comprises the step of correlating by computing a plurality of individual intra-block correlation values from which a maximum value forming a block sub-correlation value is determined, and determining a minimum 30 between the different block sub-correlations. 'and the identification information 40, 40'.

As a matter of correlation interval, in a preferred embodiment, these individual intra-block correlation values are determined by comparing the identification image 15 at the position corresponding to the block 30, 30 'with the block 40,40' of the picture and its mutual environment.

In a preferred embodiment, the method is such that, before correlation determination, one or more of the following qualifying methods are performed: classifying the identification information, subtracting the identification information and the comparison model, and providing the code. These delimitation methods reduce the need for 20 computations and speed up recognition.

Of the delimitation methods, the latter, i.e. code entry, is a per se known procedure using a personal numeric or alphanumeric code, whereby the system compares the identification image only with the corresponding reference model image according to the code. The method then performs the detection quickly.

The former, ie classification, means that reference and identification images can also be classified into different subcategories, for example on the basis of the best features of the image (consecutive and / or overlapping blocks, ie standard deviations of windows, maxima and minima, etc.). In this case, the method compares the identification image only with the same type of reference pattern images.

. 30 By subtracting the identification information from the delimiter methods and the comparison pattern, it means that the comparison image and the selected line of the identification image (or the average of the identification lines) can be subtracted. Before subtraction, it is most advantageous to perform the standardization so that the mean values of the grayscales of the rows are substantially the same, so that only the clarity and position of the image features affect the final result, for example illuminated temporal variations. The absolute (positive) values of this difference are summed. The lower the sum total, the more the images resemble each other. For example, in Figure 8 there is a difference having an average value of 7.90, which is interpreted as small, i.e., the model 30 of the reference 5 and the identification image 40 are of the same type.

The comparison of the strong features mentioned as the third pruning method means that the location of the strongest or best features of the reference image 30 and the identification image 40 can be compared. If the features are close to each other, the pictures resemble each other.

In a preferred embodiment, the correlation calculation for the reference model 30 in question is interrupted if any of the sub-correlations falls below a set threshold. For example, this limit may be 20%. This method also speeds up computation.

The identification image 40 and the reference model image 30 are substantially longitudinal to the finger 13 15 when measured in areas 16, 17, 18. The identification image 40 and the reference model image 30 would be transverse to the finger 13 if the nail line 19a is to be measured. Longitudinal or transverse images are easy to form and manipulate.

Although the invention has been described above with reference to the examples of the accompanying drawings 20, it is to be understood that the invention is not limited thereto but can be modified in many ways within the scope of the inventive idea set forth in the appended claims.

• ·.

Claims (10)

1. Biometric identification method in which identification information (40) is compiled from one or more fingers (13) in a human hand (6) and the identification information is compared with one or more previously known comparison models (30) and an identification is made if the identification information corresponds the comparison model, wherein the comparison models (30) are compiled from the same places as the identification information, characterized in that the identification information is compiled from one or more of the following mentioned places on the side of the back of a finger: finger joint folds (16, 17), folds between the finger joints (18), the nail lines (19a). A method according to claim 1, characterized in that a correlation determination is carried out which is carried out sector by sector, where the comparison model and the identification information are compared with each other sector by sector (m1, m2, m3, ... mn) and the partial correlations in the respective sector are determined, from which the output of the correlation between the comparison model (30) and the identification information (40) is determined. 3. A method according to claim 2, characterized in that the correlation is carried out so that several individual correlation values are calculated within a sector, of which a maximum value is determined, which forms the value of the partial correlation for the sector, and that of the partial correlations in the different sectors, a minimum, which is the result of the correlation between the comparison model and the identification information. 4. A method according to claim 3, characterized in that the individual correlation values within a sector are determined by comparing the sector of the comparison image (30) with the sector of the identification image (40) for the corresponding location and the surroundings on both sides of it. 5. A method according to claim 2 or 3, characterized in that the correlation is also overlapping, with the next sector of the comparison model. 30, the prevailing sector overlaps in the correlation by sector. 6. A method according to claim 5, characterized in that the overlap of the adjacent sectors is at least 50% of the length of the sector, advantageously about 75%. 7. A method according to any of the preceding claims, characterized in that the identification information and the comparison models are image information formed by the identification image and the comparison model images, the identification image comprises several picture rows and the comparison images comprise at least one image. Method according to claims 2 and 7, characterized in that in the initial stage of the correlation per sector it is determined which image row in the identification image (40) matches sufficiently the comparison image (30) and that the number of image rows in the identification image (40) with which the comparison image (30) will continue to be drastically reduced on the basis of the information received. 10 Method according to claim 2, characterized in that the correlation calculation for the part of the comparative model (30) is discontinued if any of the partial correlations is below a set limit value. Method according to claim 2, characterized in that one or more of the following thinning methods are performed before the correlation determination: classification of the identification information (40), the identification information (40) and the comparison model (30) are reduced from each other, comparison of strong ones. drag, the output of a code. ♦ · • ·