EP1437691B1 - Method for recognizing the relief of a coin in a coin operated apparatus - Google Patents

Description

The present invention relates to a method for detecting an embossed image a coin in a slot machine.

Coin machines such as coin validators discriminate a given Set of coins in a very short time. These are a series of procedures a number of which are the coinage material as distinguishing criteria starts. The thickness and diameter of the coin are also used for differentiation. However, it is not excluded for the world monetary system that for different coins the same or almost identical blanks are used. An important distinguishing feature for the coins is therefore the embossed image.

From DE 37 39 239 C2 is a method and an apparatus for processing a known embossed card. The embossed side of the card is made of two opposite, oblique directions alternately lit. The card will be different at these Illuminations recorded. The difference between the pictures taken is compared with reference data to recognize the embossed characters. This Procedure is not for highly reflective metallic surfaces of coins suitable.

From DE 33 05 509 an optical Münzprüfeinrichtung is known in which a lower An angle illuminated surface is observed at different angles. The quotient of the brightness at different angles provides information the gloss of the coin enclosed here.

From US 5,839,563 a coin validator is known in which for the recorded image a coin is a pattern comparison is performed. To get a good illumination To achieve the object field, the coin is illuminated circular.

From DE 100 51 009 a method for detecting an embossed image of a coin in which the coin moves past the light source and over two or more lighting sections each illuminated from a different direction becomes. From the recordings, a difference image is determined, indicating whether it is is a photographic reproduction of the embossed image or an embossed image.

The present invention is based on the object, a method for detection to provide an embossed image of a coin that is suitable, with simple Means in a slot machine reliably a real or false signal for the Recognize inserted coin.

According to the invention, the object is achieved by a method having the features Claim 1 solved. Advantageous designs form the subject of the dependent claims.

In the inventive method, the coin to be recognized on a Image receiver and a light source moved past. The light source has at least two, preferably three lighting sections that an object field to recognizing coin from different directions, at the same angle opposite the surface normal of the object field, not overlapping with each other Illuminated wavelength ranges. Object field of the coin is preferred the entire embossed picture of the coin. The illumination of the embossed image is done for the Lighting sections at an equal angle. The lighting led to the name "Selective Stereo Gradient Method" (SSGM), because at a central Record the picture only light at a certain angle of reflection or gradient is recorded in the embossed image. From in the inventive Procedure illuminated surface draws an image receiver a shot on. From the one shot are pictures of the individual lighting sections won. For this, the color components of the different Wavelength ranges separated. The images thus obtained become a maximum image determines at each pixel each of the maximum intensity value is assigned from the pictures. The real or false signal for the recorded Embossed image is determined from the maximum image. In the inventive Method is made up of an image, by separating it into sub-images that are made different directions, but with the same angle of inclination (azimuthal angle), determines a maximum image that the coin surface for recognition of the embossed image sufficiently well reproduces.

For the generation of the real or false signal, it has proven to be expedient for the picture and / or the maximum picture, center and diameter to determine the coin. Preferably, one or more are in the maximum image Circular ring segments cut out with predetermined radii.

Surprisingly, it has been found that for the circular ring segments of mean gray value and / or its deviation, preferably standard deviation already a good indicator of the type of coin to be recognized. Especially advantageous to this feature of the embossed image is that it comparatively low computational effort can be determined.

In a further preferred step, the values of the pixels in the maximum image along circular profile with a predetermined radius in one Frequency representation transformed. As a transformation, a Fourier transformation, which is preferably designed as a fast Fourier transform (FFT) is proved to be suitable. The transformed spectra are using reference spectra compared with the deviation in the determination of the real or False signal is taken into account. Also in this process step has surprisingly pointed out that already the spectral comparison along Circular profile sufficient to reliably an indication of real or falsehood of the embossed picture to win.

In a further preferred embodiment, in each case pairs of images are added determines the individual lighting sections differential images. The difference pictures on the one hand allow discrimination of photographic reproductions of the Embossed and on the other hand are particularly suitable to individual cutouts from the difference images with reference patterns to a match to compare, so-called template matching. In this process will be Excerpts from images of an embossed image compared with reference images.

In the method according to the invention, a false signal is generated when the average gray value of a difference image is below a predetermined threshold.

The separation of the images from the image is achieved by the use of filters, which are permeable to the individual wavelength ranges reached. Prefers When using a CMOS or CCD camera, mosaic filters are used.

In the method according to the invention, a first step is followed by a first step Classification of possible types of coins, taking into account the possible coin types First, those are excluded where starting from the maximum picture the mean gray value and / or the deviation outside of a predetermined one Interval lie; for the remaining coin types are transformed Spectra with the characteristic frequencies of the reference spectra compared.

In one development, after comparing the spectra one or more Difference pictures determined and fragmentary with reference patterns still to compared to comparative coin types.

Preferably, a real signal for the coin to be tested is always generated when reduces the number of possible coin types to a single possible coin type has been. Preferably, the false signal is always generated when no Coin type is more possible.

Fig. 1

ein Flußdiagramm für einen Münzprüfer,

Fig. 2

Nassie-Schneidermann-Diagramm für eine beispielhafte Anwendung, bei der Euromünzen aus unterschiedlichen Ländern diskriminiert werden, und

Fig. 3

ein Beispiel für ein in einem Unwarp-Bild erfolgreich gefundenes Referenztemplate.

The method according to the invention will be described in more detail below in a preferred embodiment. It shows:

Fig. 1

a flowchart for a coin validator,

Fig. 2

Nassie-Schneidermann diagram for an exemplary application in which euro coins from different countries are discriminated, and

Fig. 3

An example of a reference template successfully found in an Unwarp image.

The 3-color SSGM (3-Color Selective Stereo Gradient Method) consists of three Special partial images of the coin are the basis for the evaluation of the topography the embossed pictures. The partial images are extracted from a single image.

Here, a LED color lighting ring with five LEDs of the colors red, blue and green separated by colors into three 120-degree sectors. While running the coin is triggered by a light barrier, a trigger signal both a LED flash for all three colors simultaneously triggers, as well as the camera caused to take a single picture. The used for recording CMOS or CCD camera is with a mosaic filter, such as a Bayer Patter equipped the information from the three sectors in the recording separates. After the separation, there are again three partial images that are too illuminate the recognizable embossed image from different directions. In the flowchart from Fig. 1, the image recordings in method step 10 is shown and is separated into three individual images in method step 12. This is on the Colors red, green and blue used.

From the partial images, a differential image is calculated in step 14, the use of which ensures the security against counterfeiting of photos. In the method according to the invention, the difference image is used only for embossing image verification; the classification is subsequently carried out on the clearly structured maximum image 16. The maximum image M is defined as M (x, y) = max [R (x, y); G (x, y); B (x, y)], wherein for each point (x, y), the value of greatest intensity is selected from the set of images R (x, y), G (x, y) and B (x, y).

In the subsequent method step 18 is recorded in originally Picture the coin searched, cut out and in a picture format with predefined Image size, for example, 256 x 256 pixels brought. This operation will also the diameter determined. In this step, the images are independent of the Coin size scales to the same size, which will be discussed below Comparison of the mean gray values is important.

In step 20, a segmentation of the image takes place, starting from the image center, Circular regions are considered. It has turned out that the Disassembly of the image into an outer ring, a middle ring and a coin center is particularly advantageous for 2 euro coins.

In the following step 24, the average gray value for the three ring regions and the standard deviation of the gray values in the three ring regions certainly.

In addition, in the outer ring of the coin is another distinguishing feature compared. For this purpose, the outer ring is transformed into a binary image by using appropriate Thresholds converted. The binary images are perpendicular to each other on two standing axis projected. A characteristic feature of the embossed image The coin is the distance between the center-of-gravity (COG) and the geometric one Center of the picture. The center-of-gravity (COG) is considered to be with the Distance of weighted mean of the pixels determined. The picture is here scaled image in which the outer ring is considered. The COG is about the Axis projection method determined. Instead of a binary image can also be Gray value or color image can be used.

In a further method step 22, which is parallel to that described above Process step 24 may take place, but which is also feasible downstream, be on circular rings around the center of the scaled images gray scale profiles sampled. The radii of the circular rings are predetermined here. The values of the pixels along the circular ring profiles are Fourier transformed (FFT). The dominant Frequency of each FFT spectrum is determined. The for the five circular rings The dominant frequency determined in its entirety constitutes another characteristic feature Feature for the coin.

In step 26, a pre-comparison is performed, whether the previously obtained Measured values for the recorded coins at predetermined admissibility intervals lie. This comparison results in a classification in step 28. Represents in the classification step 28 that the pattern with none of the predetermined References match, then the classification process leads to rejection 30 of the coin. If it turns out that several coins come into question, then in Step 32 performed a template match for these coin types. As shown in Fig. 3, this is the recorded image of the coin unwound and at the double angle range to avoid cuts in the Reference pattern added. The thus supplemented image 34 is provided with a reference pattern 36 compared. As shown in FIG. 3, for the reference pattern 36, the position in FIG Image 34 found.

The method step 32 is a verification step in which it is checked whether a Coin type in the shortlist actually comes into consideration. It is conceivable here also a modification of the method step such that every possible coin type is folded with the reference images of all possible coin types to the coin type determine the highest match.

In the final step 34, for the difference image calculated in step 14 checks whether this is a stamping picture or a photo of a stamping picture is. This step can also be at the beginning of the comparison.

For this step, a variety of different approaches can be selected become. In practice, two approaches have proven to be particularly advantageous. In a first approach, the size of the template pattern at the location in the image, in the pattern was found, cut out a picture detail in size of the template. This image detail is created by the application of thresholds in a binary image changed. For example, to set the threshold value Sum of mean gray value in the pattern plus the standard deviation of the gray values be set in the pattern. Other variable or fixed thresholds are conceivable.

From the three partial images difference images are determined. As a particularly advantageous has proved here:

To obtain a first difference image (Diff1) from the images for red (R) and green (G). A second difference image (Diff2) is taken from the images for red (R) and blue (B) determined. Finally, a difference image (Diff12) as a difference between formed the first and the second difference image. However, it can also be a maximum picture between the first difference image (Diff1) and the second difference image (Diff2) are formed: Diff12 = max [Diff1, Diff2]. From the finished Difference image (Diff12) becomes an unfolded and double angle range completed picture made. This picture is congruent with the one before Image produced for the pattern comparison. Out of these unfolded images becomes the same place a picture taken in size of the reference pattern. The extracted image sections are multiplied together and for the product the mean gray value is calculated. Is the average gray value below one predetermined fixed threshold, it is a photo. It remains in the Case of a photo after multiplication of the original image with the difference images not enough information is left in the gray values of the product image.

As an alternative to the aforementioned operation, the reference template can also be used be transformed into a binary image at an appropriate threshold. For example can again be the sum of the average gray value in the pattern plus the Standard deviation of the gray values in the pattern can be used. Others too variable or fixed thresholds are possible. It will, as described above, determines the difference images, again the two image sections, the difference value and the binary reference image, are multiplied together and used for the Product, the average gray value is calculated. Again, a missing three-dimensional Topology recognized that the average gray value below a predetermined source lies.

Other methods for combining the difference images with sub-images are possible. In the above two variants, the basic idea is common, through the combination of partial images to produce differential images that only then have information content in the form of gray value structures, if that too recognizing coin has an embossed image. Should the coin recognition through a photo be deceived, the differential images do not have sufficient information. The key to capture the information of the coin's 3D imprinted image lies in the lighting. The lighting must be homogeneous and in the sectors have the same intensity.

After successful embossing image recognition, the recognized coin in step 36 accepted.

2 shows a structured diagram of the sequence of the invention Method on the example of a detection of coins of different countries.

1. Mittlerer Grauwert im Ring 1,

2. Standardabweichung der Grauwerte im Ring 1,

3. Mittlerer Grauwert in Ring 2,

4. Verhältnis der mittleren Grauwerte in Ring 1 und 2,

5. Standardabweichung der Grauwerte in Ring 2,

6. Mittlerer Grauwert in Ring 3,

7. Verhältnis der mittleren Grauwerte von Ring 2 und Ring 3,

8. Standardabweichung der Grauwerte in Ring 3,

9. Verhältnis der mittleren Grauwerte von Ring 1 und 3,

10. Abstand des COG im Ring 1 zum geometrischen Mittelpunkt der Münze bzw. zum Zentrum des Rings,

11. Durchmesser,

12.-16. Dominante Frequenzen der Grauwerte auf fünf Kreisringen.

In step 28, three color sub-images (R, G, 13 images) are extracted from the captured camera image of the coin. In step 40, a maximum image is calculated from the three partial images. In step 42, the coin is segmented from the captured image and the segmented image is rescaled to a square format for further processing. In step 44, the two steps described above are performed, in which average gray values and dominant frequencies are compared with reference images. So, not all countries of the class under review have been fully tested, loop 46 is repeated in the procedure. Overall, the following characteristics are present:

1. Middle gray value in ring 1,

2. standard deviation of the gray values in ring 1,

3. Middle gray value in ring 2,

4. Ratio of mean gray values in ring 1 and 2,

5. standard deviation of the gray values in ring 2,

6. Middle gray value in ring 3,

7. Ratio of the mean gray values of ring 2 and ring 3,

8. standard deviation of the gray values in ring 3,

9. ratio of mean gray values of ring 1 and 3,

10. distance of the COG in the ring 1 to the geometric center of the coin or to the center of the ring,

11. diameter,

12th-16th Dominant frequencies of gray values on five circles.

On the basis of the abovementioned measured values or a subgroup of these measured values a comparison is now made in step 48. If the comparison shows that the Measured values lie in a predetermined admissibility interval, so comes the corresponding Coin type shortlisted 50. Only in the frequency comparison If no uniform admissibility interval is specified, but for the Frequency three possible measured values allowed. Each of the three measured values is with one Admittance interval linked. A dominant frequency is detected when the measured frequency within one of the three before certain admissibility intervals lies.

In step 54 it is checked whether at least one type is shortlisted. In this case, in step 56, two unfolded images are calculated, one of which first from the image scaled in step 42, and the second from the reference image is.

As long as unchecked coin types are available, the following classification loop will be used executed. In step 60, for the type of coin to be checked the corresponding reference template (reference pattern) is loaded. In step 62 the pattern is compared with one of the unfolded images generated in step 56. If the template is found in the unfolded image and exceeds the match a minimum value (step 64), then in step 66 the location of the template into the unfolded image. For the cutouts, the above described comparison producing the product image in step 68 carried out. Subsequently, the stamp image verification is performed. Is the Value for the embossing image verification bad, so in step 72 the tested one Coin also off. Should after the completion of the loop 58 more than one coin If left over, these can be the coins with the best comparative values (for example, from step 48).

Claims (16)

1. A process for identifying an embossed image of a coin in an automatic coin machine, comprising the process steps below:

   The coin requiring identification is moved past an image receiver and a light source,

   The light source has at least two lighting portions which illuminate an object field of the coin requiring identification from different directions under the same angle with respect to the surface normal of the object field with wavelength ranges which do not overlap each other,

   The image receiver records one picked-up exposure of the object field from which images are gained for the individual lighting portions of the individual wavelength ranges,

   A maximum image is determined from the images wherein each pixel has associated therewith the maximal intensity value from the images of the individual wavelength ranges,

   A genuine-coin or counterfeit-coin signal is determined from the maximum image.
2. The process as claimed in claim 1, characterized in that the centre and diameter of the coin are determined for the picked-up exposure and/or the maximum image.
3. The process as claimed in claim 2, characterized in that one or more circular ring segments having predetermined radii are cut out in the maximum image.
4. The process as claimed in claim 3, characterized in that the genuine-coin or counterfeit-coin signal is generated using the mean grey-scale value of the circular ring segments and/or a deviation of the grey-scale value from the mean grey-scale value.
5. The process as claimed in any one of claims 1 to 4, characterized in that the values of the pixels in the maximum image are transformed into a frequency representation along circular ring profiles having a predetermined radius.
6. The process as claimed in claim 5, characterized in that a Fourier transform, specifically a fast Fourier transform (FFT), is carried out as a transformation of the values of the pixels.
7. The process as claimed in claim 5 or 6, characterized in that the transformed spectra are compared to reference spectra and the deviations are taken into account in determining the genuine-coin or counterfeit-coin signals.
8. The process as claimed in any one of claims 1 to 7, characterized in that differential images are determined for the individual lighting portions from pairs of images.
9. The process as claimed in claim 8, characterized in that one or more sections from the differential images are compared to reference patterns for their coincidence.
10. The process as claimed in any one of claims 1 to 9, characterized in that the separation of the picked-up exposure into images of the individual lighting portions is performed via filters.
11. The process as claimed in claim 10, characterized in that a mosaic filter for a CMOS or CCD camera is used for the separation of the exposure into images.
12. The process as claimed in any one of claims 1 to 11, characterized in that a counterfeit-coin signal is generated when the mean grey-scale value of a differential image is below a predetermined threshold value.
13. The process as claimed in any one of the preceding claims, characterized in that

    a classification is initially made for possible coin types in a first step wherein those of the possible coin types are initially excluded for which the mean grey-scale value and/or the deviation, based on the maximum image, are outside of a predetermined interval and/or the transformed spectra with their characteristic frequencies deviate from predetermined reference spectra of the coin types.
14. The process as claimed in claim 13, characterized in that one or more differential images are determined for the coin types still remaining after a comparison, from the exposures of the wavelength ranges and are compared by sections to reference patterns of the coin type yet to be compared.
15. The process as claimed in claim 13 or 14, characterized in that a genuine-coin signal is generated whenever the number of the possible coin types was reduced to a single coin type.
16. The process as claimed in claim 15, characterized in that a counterfeit-coin signal is generated whenever the number of the possible coin types was reduced to zero.

Images