EP1897067B1 - Method and device for recognising a coin by using the embossed pattern thereof - Google Patents

Abstract

The invention relates to a method and device for recognizing a coin by using the embossed pattern characteristics thereof. For this purpose, the inventive method comprises in spreading the characteristics of the picture, in reducing said characteristics by reducing said picture and in transforming it by polar transformation, in comparing the transformed picture with a plurality of reference patterns according to a first simplified criterion, in creating a list of the reference patterns, in sorting them according to the similarity thereof with the transformed picture and in comparing the transformed picture with the reference patterns contained in the list according to the sorting thereof upon a second exact criterion.

Description

The invention relates to a method for detecting a coin entered into a coin validator using its embossed image according to the preamble of the main claim and an apparatus for performing the method.

From the DE 102 02 383 A1 For example, there is known a method of detecting an embossed image of a coin in a coin acceptor, wherein an image receiver receives an image of the embossed image of the coin moved to the image receiver and to a light source. An evaluation device compares the image with a first reference pattern as to whether the first reference pattern is contained in the recorded image, and if it is contained, it is checked whether a second reference pattern is contained in an area whose position is determined relative to the position of the first reference pattern. The evaluation device generates a real or false signal for the coin depending on the coincidence of the image with the reference patterns. During the evaluation, the center point is determined for the recorded image, and furthermore the image is transformed into circle coordinates, whereby the transformed image is the basis for the search of the reference patterns.

Furthermore, in the EP 0 798 670 B1 a method for detecting the embossed image of a coin is described in which again the image of the coin is taken, the center determined and a polar transformation is performed. At a predetermined distance from the abscissa in the polar transformed image, the transformed embossing image is scanned and compared with a reference pattern at an appropriate pitch, the patterns being translated relative to each other to angularly align the measured coin with the reference coin.

From the US 2004/0168881 A1 For example, there is known a method for recognizing the embossed image of a coin which is illuminated by a light source having different wavelengths in different illumination sections and recorded by an image sensor as a set of images. From the set of images, a maximum image is determined by assigning to each pixel of the maximum image the value with the maximum intensity from the set of images. A true or false signal is generated from the maximum image using an average gray value of circular ring segments with predetermined radii cut out of the maximum image.

One of the main problems with the embossing image analysis is this large amount of data that must be processed in the time that the coin falls through the machine to ensure reliable detection. In order to be able to measure the diameter with an accuracy of, for example, 0.1 mm, the overall image of the coin must have a resolution of at least 100 pixels per mm. An average coin of about 20 mm in diameter is then imaged with 200 x 200 pixels. Even if only a relatively large fragment of the coin surface is selected for comparison, the computation volumes are so large that they can hardly be carried out promptly during the insertion of the coin into a slot machine.

The invention is therefore an object of the invention to provide a method for detecting a coin entered into a coin validator using its embossed image, which allows the detection of the coin quickly and reliably.

The object is achieved by the characterizing features of the main claim in connection with the features of the preamble.

By spreading the features of a pattern layout in the image of the coin and reducing features by reducing the size of the image by subjecting the image to polar coordinate transformation, the speed of comparing the coins with reference patterns can be increased and the possibility is allowed to not only fragments, but practically the entire surface of the coin to be used as a reference pattern, which in turn increases the robustness of the process against accidental damage and contamination of the coin, the spreading of the features robustness of the comparison between the current image and a reference image, especially during shifts or twists of the coin increases. The polar coordinate transformation thereby converts the rotation of the current coin image or of the reference pattern into a linear, e.g. horizontal translation, which is much faster to calculate.

By further making a two-step comparison in which the image of the coin is compared with the reference patterns according to a first simplified criterion and a list of selected reference patterns sorted according to their

Similarity is created and then a comparison of the mapping with the reference patterns contained in the list according to their sorting according to a second, accurate criterion is made, the processing time is significantly shortened.

The measures specified in the dependent claims advantageous refinements and improvements are possible.

According to the invention, the spreading is in direct connection with the reduction, wherein the features are spread before the reduction or simultaneously with the reduction. The size of the maximum filter is determined by the reduction factor. The spread preserves the physical features of the image as it is reduced, thereby reducing the mathematical features to speed detection.

It is particularly advantageous, as a characteristic for the first simplified criterion, to calculate the distribution of the average brightness in the lines of the transformed image and then to use a one-dimensional correlation between the brightness distribution of the transformed image and the reference images or patterns. In this way, a good selection of the candidate reference pattern is already achieved in the first comparison. Based on the first simplified criterion, a list of the reference patterns is created according to their similarity to the current image.

Preferably, as a second exact criterion, a two-dimensional correlation of the brightness distribution in the transformed image can be used. An exact comparison is carried out, whereby the results of the preliminary analysis are no longer taken into account and only the result of the exact comparison counts.

Fig. 1

eine Darstellung zur Polartransformation ei- ner Münze,

Fig. 2

das Originalprägebild einer Münze, sowie zwei Polartransformationen des Prägebildes der Münze mit Reduktion der Merkmale, win- kelmäßig um 3° verdreht,

Fig. 3

Ansichten entsprechend Fig. 2, bei denen ei- ne Spreizung der Merkmale mit einem Maximum- filter vorgenommen wurde, und

Fig. 4

die Darstellung eines Verfahrensablaufs zur Auswertung des Prägebildes einer Münze in einem Münzautomaten.

Embodiments of the method according to the invention will be explained in more detail in the following description with reference to the accompanying drawings. Show it:

Fig. 1

a representation for the polar transformation of a coin,

Fig. 2

the original embossed image of a coin, as well as two polar transformations of the embossed image of the coin with reduction of the features, angularly rotated by 3 °,

Fig. 3

Views accordingly Fig. 2 in which a spread of the characteristics was carried out with a maximum filter, and

Fig. 4

the representation of a procedure for evaluating the embossed image of a coin in a slot machine.

The method according to the invention is used to identify a coin with evaluation of its embossed image. The coin is thrown into the coin validator and the image of the coin is recorded by means of an image sensor and sent as pixel data to the evaluation unit. Among other things, this evaluation unit determines the exact diameter and the exact center as well as possibly the shape. In the further evaluation, among other things, a polar transformation is done accordingly Fig. 1 performed at For example, assuming the radius of the coin as the outer radius of the transformation and the inner radius of the transformation is 0. The angle θ is counted clockwise starting from the positive x-axis. How out Fig. 1 below, creates a "distorted" pattern that can be evaluated linearly.

In Fig. 2a1 Fig. 12 is an illustration of a coin obtained in an image pickup module with illumination oblique to the coin surface, showing thin light lines on a dark background on the coin surface. These thin lines represent characteristic features of the coin that make up a pattern, or portions thereof. In order to increase the speed in the later evaluation, ie in the comparison with reference features or patterns, it is advantageous to reduce the number of features. The reduction of the features could be performed, for example, by reducing the size of the image by subsampling pixels. For a reduction factor N, each line of the original image is processed only every Nth pixel, all others are omitted. The same applies to undersampling of picture lines. In such a sub-scan, a part of the features contained in the original image is lost. With a slight rotation or displacement of the original image, different features always remain, and the corresponding transformed images are dissimilar.

In Fig. 2a2 and a3 is a polar transformation accordingly Fig. 1 shown in which a so-called subsampling made directly during the transformation ie, the image was transformed with a reduction or reduction factor N, eg 6. The transformed images accordingly Fig. 2a2 and a3 are in proportion to Fig. 2a1 shown magnified, wherein the coin at a3 gegenübe4r a2 was rotated rotated by 3 ° and the two images based on the same transformation. It has been shown that in this treatment accordingly Fig. 2a2, a3 it is likely that by omitting pixels, features will also be omitted, thereby reducing visibility.

In order to avoid the uncontrolled loss of information in the reduction of the features by sub-sampling, a spread of the image is made, the result of the spread in Fig. 3 is shown. With the spreading, a physical enlargement of the characteristic features is carried out in each case on several pixels.

The spreading can be carried out in different ways, in an image, such as in Fig. 2a1 is shown having bright lines on a dark background, a spread of the features, ie the bright lines by filtering with a maximum filter is feasible. This is in Fig. 3b1 in which it can be seen that the "bright" features are physically enlarged and distributed over several pixels.

If the image of the coin is taken with a vertical illumination in the recording module, dark lines on a light background can be seen in the image, in this case the spreading can be carried out for example by filtering with a minimum filter.

To make the image smaller, the size of both the maximum and minimum filters is defined as NxM pixels, where N and M correspond to the reduction factors along the columns and rows. Subsequently or simultaneously with the filtering and reduction, which are based on a processing of the pixels, the polar transformation can accordingly Fig. 3 Figure 2.

Fig. 3b3 is a representation accordingly Fig. 2a3 , in which the embossed image by 3 ° to the representations of the FIGS. 2a2 and 3b2 is twisted. As can be clearly seen, the features occur accordingly Fig. 3 and the similarity between the figures b2 and b3 is much higher than that between the figures even after the correlation measure calculated later Fig. 2a2 and a3 , In this way, first the spreading and then the reduction or transformation is carried out with a reduction.

In another embodiment of the spreading of the features this is achieved with a modified polar transformation, at the same time the image is reduced. For this purpose, to a point in the transformed image with the Cartesian coordinates θ, r, a corresponding origin point in the original image is calculated with distance from the center of the coin N * r and an angle of M * θ against an orientation determined for the image and the brightness of the point in the transformed image is calculated as the maximum of the brightness of the original image on a region of size K * K pixels around the origin point, where K is the maximum of the reduction factors: K = max (N, M).

With this method of spreading and reduction by the modified polar transformation, using Fig. 2a1 achieved the same results as in Fig. 3b2 and Fig. 3b3 are shown.

After the spreading, reduction and polar transformation, which can take place simultaneously as described above, a multi-level comparison of the transformed image corresponding Fig. 3b2 and b3 performed with a number of reference patterns. For this purpose, a first simplified criterion is used as the basis for the transformed reduced image with spread characteristics in the first stage, although it can not reliably identify the coin, but requires only a short processing time. The comparison of the transformed map with all the reference patterns on the basis of the first simplified criterion results in each case a similarity measure with which a sorted, temporary list of the reference patterns is created. Patterns that provide better results, ie greater similarities, are positioned at the top of the list. As a result, in a comparison in a second stage, the matching reference pattern can be found with a high probability among the first, present in the list candidate, whereby the processing time is significantly shortened.

As a characteristic for a simplified criterion, the distribution of average brightness in lines of the transformed image and, as a simplified criterion, a one-dimensional correlation between these characteristics for the transformed image and the reference patterns may be used become.

In the second stage, a second comparison is made between the transformed map and the reference samples on the list according to a second, more accurate criterion requiring a greater processing time. In this case, a match with good accuracy is found with one of the reference patterns and a signal for the admissibility of the coin is given or the process of comparison is aborted. As a characteristic for the second, exactly criterion, e.g. serve the two-dimensional brightness distribution in the transformed image, and the comparison can be carried out for example by means of the two-dimensional correlation.

Since there is only a predetermined amount of time to check the coin in the coin validator, the check must be canceled and the coin returned when the time is up. For example, the actual comparison process can be aborted after a predetermined number of reference patterns corresponding to the given list. The maximum number of reference patterns to be processed can be determined depending on the capacity of the computer. Another possibility is to perform the comparison calculations of the reference patterns according to their sorted order until the coin comes in its course through the coin validator to a predetermined location, for example, to the point at which it is sorted. If there is still no valid classification result at this time, the coin falls into the return shaft.

It is possible that after this second comparison stage, a final decision can already be taken on the acceptance or rejection of the coin, in particular if all reference patterns defined by coin classes are easily separable. To a class of coins, e.g. in each case all valid coins with the same denomination are assigned. Then each class will consist of at least two reference patterns, a pattern for a header page and a pattern for a pay-page. If there are several valid variants for header images of the header or the payment page, the number of patterns becomes higher. Nevertheless, apart from intentional forgeries, all embossed images are usually so different that high correlation quotients are only possible between images of a class.

Another situation occurs when the similarity of the embossing or the transformed image to one of the reference patterns of the coin class X is indeed established, but a final recognition can not be performed because there are also other coin classes, their similarity with the coin class X before is known. For example, This concerns counterfeiting of coins which, in the case of 'good forgery', can very much resemble genuine coins. It can not be ruled out that with the same diameter and similar imprints sometimes the real coins can also have different denominations. In this case, an additional safety check as a third stage is necessary for a final decision.

wobei h_i(x,y) und h_j(x,y) mittelwertfreie Helligkeitsverteilungen in den Referenzmustern der ähnlichen Klassen i und j sind. K ist ein Faktor, der so ermittelt werden kann, dass nur deutlich unterschiedliche Stellen mit einbezogen werden, beispielsweise: $$\mathrm{K}\left(\mathrm{x},\mathrm{y}\right)\mathrm{=}\left\{\begin{array}{l}\mathrm{1}\mathrm{\Leftarrow }{\mathrm{h}}_{\mathrm{i}}\left(\mathrm{x},\mathrm{y}\right)\mathrm{<}\mathrm{0}\mathrm{,}\mathrm{5}{\mathrm{h}}_{\mathrm{j}}\left(\mathrm{x},\mathrm{y}\right)\mathrm{;}{\mathrm{h}}_{\mathrm{i}}\left(\mathrm{x},\mathrm{y}\right)\mathrm{>}\mathrm{2}{\mathrm{h}}_{\mathrm{j}}\left(\mathrm{x},\mathrm{y}\right)\\ \mathrm{0}\mathrm{\Leftarrow }{\mathrm{h}}_{\mathrm{i}}\left(\mathrm{x},\mathrm{y}\right)\mathrm{\ge }\mathrm{0}\mathrm{,}\mathrm{5}{\mathrm{h}}_{\mathrm{j}}\left(\mathrm{x},\mathrm{y}\right)\mathrm{\cap }\mathrm{hi}\left(\mathrm{x},\mathrm{y}\right)\mathrm{<}\mathrm{2}\mathrm{hj}\left(\mathrm{x},\mathrm{y}\right)\end{array}\right\}$$

The brightness distribution in the transformed image can also be used for the safety test. If there are differences in particular fragments of the imprints, these fragments can be patterned selected for the safety test. If different features are scattered throughout the figure, a difference characteristic of the features can be calculated as follows: $${\mathrm{U}}_{\mathrm{ij}}\left(\mathrm{x},\mathrm{y}\right)\mathrm{=}\mathrm{K}\left(\mathrm{x},\mathrm{y}\right)\mathrm{*}\left({\mathrm{H}}_{\mathrm{i}}\left(\mathrm{x},\mathrm{y}\right)\mathrm{-}{\mathrm{H}}_{\mathrm{j}}\left(\mathrm{x},\mathrm{y}\right)\right)\mathrm{.}$$

where h _i (x, y) and h _j (x, y) are mean-free brightness distributions in the reference patterns of the similar classes i and j. K is a factor that can be determined to include only significantly different locations, for example: $$\mathrm{K}\left(\mathrm{x},\mathrm{y}\right)\mathrm{=}\left\{\begin{array}{l}\mathrm{1}\mathrm{\Leftarrow }{\mathrm{H}}_{\mathrm{i}}\left(\mathrm{x},\mathrm{y}\right)\mathrm{<}\mathrm{0}\mathrm{.}\mathrm{5}{\mathrm{H}}_{\mathrm{j}}\left(\mathrm{x},\mathrm{y}\right)\mathrm{;}{\mathrm{H}}_{\mathrm{i}}\left(\mathrm{x},\mathrm{y}\right)\mathrm{>}\mathrm{2}{\mathrm{H}}_{\mathrm{j}}\left(\mathrm{x},\mathrm{y}\right)\\ \mathrm{0}\mathrm{\Leftarrow }{\mathrm{H}}_{\mathrm{i}}\left(\mathrm{x},\mathrm{y}\right)\mathrm{\ge }\mathrm{0}\mathrm{.}\mathrm{5}{\mathrm{H}}_{\mathrm{j}}\left(\mathrm{x},\mathrm{y}\right)\mathrm{\cap }\mathrm{Hi}\left(\mathrm{x},\mathrm{y}\right)\mathrm{<}\mathrm{2}\mathrm{hj}\left(\mathrm{x},\mathrm{y}\right)\end{array}\right\}$$

The comparison of this difference of a transformed map, which is more similar to the class i, gives a positive signal and a map similar to the class j gives a negative signal. If several classes are similar to each other, such differences or difference fragments must be generated and checked for each class pair.

In order to further reduce the number of misrecognitions, reference patterns of coin embossments, which may be particularly frequent at the location of the coin validator concerned, may be subjected to the detailed examination according to the second criterion independently of the test according to the simplified first criterion. This includes, for example, a number page that is identical for all euro coins and whose probability of appearance as the current figure is 0.5. These should be checked in any case. For example, such patterns may be added to the beginning of the sorted temporary list without making a comparison according to the simplified criterion.

In Fig. 4 a process flow of the method according to the invention is shown. The evaluation device of the coin validator receives from the image acquisition module a current high-resolution image of the coin with a precisely determined diameter, shape and center point. The determined diameter and the determined shape are compared in step S1 with the list of permissible diameters and the shape of the coin. If an invalid value or an invalid shape is present, the coin is rejected immediately.

In the permissible case, the image of a modified polar transformation with simultaneous spreading of the features and reduction of the image is subjected to the image in step S2, whereby the transformed image corresponding to Fig. 3b2 or b3 results. Of all the reference patterns stored in the system, those for the embossing image recognition belonging to a coin of the same diameter are selected. For the transformed image, a characteristic for the simplified criterion is calculated in step S3, for example a distribution of the average brightness for the individual lines of the transformed image. This characteristic is compared in step S4 with the corresponding characteristics of the reference patterns stored for the current diameter in a database PKRM, wherein all the patterns are sorted in the order of decreasing similarity. Thus, a temporary sorted list of the reference patterns is formed (see S5). There is an additionally stored frequency list HL. If the reference patterns of the list appear in this frequency list, they become Added pattern at the top of the list without comparing their characteristics.

The current transformed image of the coin is compared in step S6 with the first reference pattern from the temporary list according to the second, accurate criterion according to a two-dimensional brightness distribution, e.g. with the help of a two-dimensional correlation. For this purpose, the corresponding reference patterns are supplied from the database GKRM. If it is determined in step S7 that the result of the comparison with the respective reference pattern A of the class X exceeds a predetermined degree of similarity x, the comparison is aborted and the coin is provisionally sorted into a class X. If the class X has no known similarity to another class, this preliminary classification is confirmed and the procedure is terminated, i.e., the coin is recognized as permissible.

If it is determined at step S7 that the similarity to the treated reference pattern is not large enough, it is determined in step S8 whether there is another reference pattern in the temporary list TLRM. If so, the process returns to step S6 and a retest begins.

If it is determined in step S7 that there is a possibility of being confused with a reference pattern of a class Y, the security check is carried out in step S9, for example searching for either fragments that occur in one of the classes and not in another or The current transformed image is compared with a differential characteristic. The reference patterns or the reference values for the security check are stored in a database MSP.

The comparison of the current embossed image or the transformed image with the reference patterns is aborted after the predetermined time if there is no valid result.

The evaluation unit, i.e. the calculation, comparison and storage means, may be in the form of a microprocessor, microcomputer or the like with corresponding memories as stated above.

In the above embodiment, as a "simplified criterion" for the comparison, the result of one-dimensional correlation between the average brightness distributions in lines of the transformed image was used as specific characteristics. As another example of a simplified criterion, a result of a particular operation could be the distribution of the light and dark pixels in one of the lines of the transformed image. For example, in the illustration of a number page of a German coin with a denomination of 1 or 2 euros, there are more dark pixels than bright ones, and in a head page of the same coin there are more bright pixels than dark ones. If the quotient number light / number dark is used as a criterion, the head side of a German coin can be distinguished from its pay side.

Furthermore, one can use the same characteristic, namely the distribution of the average brightness in the lines of the transformed image, but choose a different criterion for comparison. For example the coordinate of the maximum of the distribution can be used. If, in the current image, the maximum lies at the edge of the coin, only the reference images that have the maximum of the distribution at the edge, etc., are selected for the exact comparison.

Claims (16)

1. Method for recognising a coin which is inserted in a coin-acceptor unit by using the embossed image thereof which has characteristic features, said embossed image being recorded by a camera device,
   characterised by the following steps:

   spreading of the features, which prescribe a pattern arrangement, in the image of the coin,

   reducing the features by reducing the image and

   transforming the same with a polar transformation,

   comparing the transformed image with a plurality of prescribed reference patterns, according to a first simplified criterion, with a rapid processing time and

   producing a list of reference patterns, sorted according to their similarity to the transformed image,

   comparing the transformed image with the reference patterns contained in the list corresponding to the sorting thereof according to a second, exact criterion and emitting a recognition signal if one of the reference patterns corresponds to the transformed image,

   stopping the comparison process and rejecting the coin according to a prescribed condition.
2. Method according to claim 1, characterised in that the spreading of the features and the reduction of the image are implemented at the same time.
3. Method according to claim 1 or claim 2, characterised in that the spreading of the features and the reduction of the image are implemented with a maximum or minimum filter, the N x N pixels of which corresponds to the reduction factor.
4. Method according to claim 3, characterised in that the reduced and spread image is subjected to a polar transformation.
5. Method according to claim 1 or claim 2, characterised in that the spreading of the features and the reduction of the image is implemented with a modified polar transformation, a corresponding origin point in the image with a spacing N*r from the centre of the coin and an angle M*θ relative to an orientation determined for the image corresponding to any point in the transformed image with the Cartesian coordinates θ, r and the brightness of the point in the transformed image being calculated as maximum of the brightness of the image on the area of the K*K pixels around the origin and K being the greatest of the reduction factors (N, M).
6. Method according to one of the claims 1 to 5, characterised in that, for the first, simplified criterion, a line-wise calculation of the brightness distribution in the transformed image is undertaken.
7. Method according to claim 6, characterised in that the comparison between the transformed image and the reference pattern is implemented with a one-dimensional correlation between the brightness distribution for the transformed image of the coin and the reference pattern.
8. Method according to one of the claims 1 to 7, characterised in that, for the second, exact criterion, a two-dimensional brightness distribution is used in the transformed image.
9. Method according to claim 8, characterised in that the comparison between the transformed image and the reference pattern is implemented with the help of a two-dimensional correlation.
10. Method according to one of the claims 1 to 9, characterised in that, as prescribed condition for stopping the comparison process according to the exact criterion, a predetermined number of reference patterns to be processed is chosen.
11. Method according to one of the claims 1 to 9, characterised in that, as prescribed condition for stopping the comparison process, a predetermined time is chosen, in particular that which the coin requires until exiting from the coin-acceptor unit.
12. Method according to one of the claims 1 to 11, characterised in that, in the case where, after the comparison corresponding to the second, exact criterion, it is established that a similarity of the transformed image to at least two coin classes is present, an accuracy test is undertaken in which features are determined for the transformed image which are different for the at least two coin classes.
13. Method according to claim 12, characterised in that a predetermined fragment of the transformed image is compared with the at least two reference patterns and the similar reference pattern is determined.
14. Method according to claim 12, characterised in that at least one difference pattern between reference patterns is produced and, by means of this comparison with the transformed image, a mostly similar reference pattern is determined.
15. Method according to one of the claims 1 to 14, characterised in that selected reference patterns are added to the beginning of the list of reference patterns which is sorted according to the result of the comparison corresponding to the first, simplified criterion of similarity, independently of this result, and are subjected to a test according to the second, exact criterion of similarity.
16. Device for implementing the method according to one of the claims 1 to 15 with a camera for recording the embossed pattern of the coin which has characteristic features and an evaluation device, characterised in that the evaluation device comprises:

    calculation means for spreading the features in the image, reducing the features by reducing the image and transforming the same with a polar transformation,

    first comparison means for comparing the transformed image with a plurality of prescribed reference patterns, according to a first simplified criterion, with a rapid processing time,

    means for producing a list of reference patterns, sorted according to their similarity to the transformed image,

    second comparison means for comparing the transformed image with the reference patterns contained in the list corresponding to the sorting thereof according to a second, exact criterion and emitting a recognition signal if one of the reference patterns corresponds to the transformed image, and

    first memory means for storing characteristics of the reference patterns for comparison according to the first criterion,

    second memory means for intermediate storage of the list produced by the means for producing a list, and

    third memory means for storing characteristics of the reference patterns for comparison according to the second criterion.

Images