EP4143733A1

EP4143733A1 - Method for verifying a barcode

Info

Publication number: EP4143733A1
Application number: EP21727776.3A
Authority: EP
Inventors: Marc Pic; Mohammed Amine OUDDAN
Original assignee: Surys SA
Current assignee: Surys SA
Priority date: 2020-04-30
Filing date: 2021-04-30
Publication date: 2023-03-08
Also published as: WO2021219852A1; FR3109831A1; FR3109831B1

Abstract

The invention relates to a method for processing a reference image of a reference barcode having random printing defects, the method comprising the steps of: - recording, in a memory, a greyscale reference image of the reference barcode, - recording an identifier (UID) of the reference barcode in the form of an index in a database. It is essentially characterised in that it further comprises the steps of: - calculating a reference fingerprint of said reference barcode, the fingerprint being a function of the random printing defects, and - recording the reference fingerprint in the database.

Description

Method for verifying a bar code.

FIELD OF THE INVENTION

The present invention relates to the field of barcode verification, including two-dimensional barcodes, whether they are Datamatrix, QR code, PDF417, 1D barcode or 2D barcode, or even a character string. , and hereinafter "bar code" or "bar code" for brevity.

Barcodes are used on multiple products, especially for traceability purposes, for example to track and trace product movements throughout a supply chain, which can detect fraudulent activity, market illicit, etc.

Depending on the case, they can be registered on a primary packaging or on a secondary packaging.

Two-dimensional (2D) barcodes include a variety of symbols (rectangles, dots, hexagons, and other geometric shapes) positioned as a matrix of juxtaposed dots or squares. These points or squares are also called modules. As described later, an image of a module is referred to as a "blob" in the present invention.

For brevity again, only the two-dimensional barcode case for the tobacco industry will be described here. In this example, the barcodes are 6 x 6 mm datamatrix.

In the context of regulatory tobacco traceability, cigarette packages are today traced in Europe by means of a unique identifier (UID) which is engraved directly on cigarette packages on production lines, at the level of an area of black paint, usually located on the bottom of the package.

In order to strengthen consumer security and the fight against counterfeit fraud, it is desirable to have a means of verifying the authenticity and uniqueness of the UID (Unique Identifier), that is to say of the uniqueness of the package with this number. For the purposes of the present invention, packet and UID are understood indiscriminately. For the purposes of the present invention, a UID is defined in particular by the ISO 22300: 2018 standard, that is to say a code which represents a unique and specific set of attributes linked to an object or to a class of objects. throughout its life in a particular domain and perimeter of an object identification system.

According to the invention, the bar code allows, on the one hand, traceability and, on the other hand, makes it possible to calculate a characteristic imprint for each package.

SUMMARY OF THE INVENTION

More precisely, the invention relates to a method of enrolling a reference two-dimensional bar code, by processing a reference image of one of said reference two-dimensional bar code, the bar code comprising a set of dots exhibiting random printing defects, each dot of the reference barcode corresponding to a respective blob in the reference image, the method comprising the steps of:

- save in a memory a reference image in gray levels of the reference barcode,

- save an identifier (UID) of the reference barcode in the form of an index in a database.

It is essentially characterized in that it further comprises steps consisting in: - calculate a reference imprint of said reference bar code, the imprint being a function of the random printing defects, and

- record the reference fingerprint in the database in the form of a vector with N components, with N the total number of fingerprint calculation methods, the position of each component in the vector corresponding to a predefined method; or as a binarized summary of the reference image such that the image of the two-dimensional barcode is stored in a compressed binary format.

In one embodiment, when the reference fingerprint in the database is recorded in the form of a vector, said vector of the reference fingerprint comprises several components, at least one of which is non-zero, each component is calculated by one of respective steps among:

- determine the area of each point of the reference bar code and record a vector which contains all the area ratios of 2 successive points of the reference bar code;

- determine the angular position of specific points located on the contour of each blob of the reference barcode;

- determine a curve characterizing the profile of each blob of the reference barcode;

- determining a set of oriented gradient histograms, calculated from the intensity and orientation of the gradients on blocks of pixels, for each point of the reference bar code;

- recording the reference image in the form of binary image comprising the coordinates of the centers of each point of the bar code; producing a curve joining the continuous points along the border of pixels having the same intensity, calculating the convex envelope of said continuous points, and calculating the convexity defects for each convex envelope; and

- produce a binary image of the reference bar code, with the coordinates of the centers of each point of said bar code, calculate the external contours of the points of said bar code, and for each contour, calculate a set of moments of the external contours of the barcode points.

According to another of its objects, the invention relates to a method for verifying a candidate bar code comprising an identifier (UID), the method comprising steps consisting in:

- record in a memory a candidate image in gray levels of the candidate barcode,

- decode the identifier (UID) of the candidate barcode,

- identify the reference bar code according to the invention, the index of which in the database is equal to said identifier of the candidate bar code,

- calculate an imprint of said candidate barcode,

comparing said imprint of the candidate barcode with the corresponding reference imprint of the reference barcode.

Provision can be made for the comparison step to include at least one of the steps among:

- a local dissimilarity map filtering step, optionally preceded by a denoising step; and

- a topological analysis of irregular points.

Steps can be provided for:

- determine the location of the candidate barcode and its orientation in a 3-dimensional space, - calculate a pose vector,

- straighten the image of the candidate barcode, and

- optionally clean said image of the candidate barcode.

It is possible to provide a step of representing the imprint of the candidate bar code and the imprint of the reference bar code by one or more sequences of bits,

the comparison step being implemented by comparing the sequences of bits which represent them.

Provision can be made for the reference bar code and the candidate bar code to be two-dimensional bar codes, the method further comprising at least one of the steps among:

- apply a frequency filter to the candidate image,

- detect the quality of the candidate image, using image quality detectors

Provision can be made for the reference bar code and the candidate bar code to be marked on a packet of tobacco, optionally covered with cellophane.

According to another of its objects, the invention relates to a computer program comprising program code instructions for executing the steps of the method according to the invention, when said program is executed on a computer.

According to another of its objects, the invention relates to a smartphone comprising a memory in which the computer program according to the invention is stored.

Other characteristics and advantages of the present invention will emerge more clearly on reading the following description given by way of illustrative and non-limiting example and made with reference to the appended figures.

DESCRIPTION OF THE DRAWINGS

[FIG. 1] illustrates an embodiment of a bar code according to the invention, and

[FIG. 2] illustrates an embodiment of a printing of the bar code of FIG. 1, exhibiting printing defects,

[FIG. 3] illustrates an enlargement of the frame of FIG. 1,

[FIG. 4] illustrates an enlargement of the frame of FIG. 2,

[FIG. 5] illustrates a candidate image comprising parasites,

[FIG. 6] illustrates a candidate image according to the invention,

[FIG. 7] illustrates the image of FIG. 6 preprocessed and filtered according to the invention,

[FIG. 8] illustrates the image of FIG. 7 binarized according to the invention,

[FIG. 9] illustrates the image of figure 8 for which the shape of the bar code is highlighted,

[FIG. 10] illustrates FIG. 9 to which an edge detection is applied,

[FIG. 11] illustrates the application according to the invention of a 2D model made up of three points on the image of FIG. 9,

[FIG. 12] illustrates clusters according to the invention, [FIG. 13] illustrates a grid generated according to the invention on the clusters of FIG. 12, with the coordinates of each blob,

[FIG. 14] illustrates a grid according to the invention applied to a datamatrix,

[FIG. 15] illustrates a table of points of a datamatrix according to the invention,

[FIG. 16] illustrates the neighborhood of points on the table of figure 15,

[FIG. 17] illustrates two identical blobs according to the invention,

[FIG. 18] illustrates two different blobs according to the invention,

[FIG. 19] illustrates two profiles of similar blobs according to the invention,

[FIG. 20] illustrates two different blob profiles according to the invention,

[FIG. 21] illustrates a topological analysis of irregular points according to 3 variants according to the invention,

[FIG. 22] illustrates the construction of filtrations and their representation in the form of a persistent bar code according to the invention,

[FIG. 23] illustrates a point cloud persistence diagram in Figure 22.

DETAILED DESCRIPTION

According to the invention, the bar codes are marked by laser engraving. This type of engraving consists in removing more or less material from the surface of the paper of the tobacco packet, which makes it possible to have more nuances to generate the barcode modules but the use of a laser generates random printing defects of the bar code, engraving and printing being considered without distinction in the present invention.

For the sake of brevity, only a two-dimensional (2D) barcode will be described here, in this case in the form of a datamatrix. According to the invention, the bar code is flat.

In the field of bar codes for tobacco packages, UIDs correspond to a construction, a structure, imposed, known.

The invention aims to characterize the printing defects of the modules of a bar code, on the basis of the assumption that these defects are never reproduced identically when the same bar code is reprinted, for example on a lot of packs of cigarettes.

Figure 1 and Figure 2 illustrate the same barcode with printing defects, boxed and enlarged in Figures 3 and 4 respectively.

The present invention provides two distinct phases: a first enrollment phase which precedes a second control or verification phase. The conditions for acquiring the images are generally different depending on the phase.

According to the invention, during the enrollment phase, before the packet is placed on the market, a so-called “reference print” of the barcode of the packet is calculated.

During the control phase, after the package has been placed on the market, a so-called “control fingerprint” of the barcode is then calculated.

We can then compare the control fingerprint and the reference fingerprint to determine if the barcode is original.

In the enrollment phase, a camera is provided on the package production line to acquire a set of reference images. The production line involves constraints of packet scrolling speed. Packages are generally cellophane free when acquiring these reference images.

In the control phase, a communicating object is provided, in particular a smartphone, to acquire candidate images. For brevity, the communicating object, that is to say any manipulable optoelectronic device comprising an optical objective, programmable means of calculations and means of wireless communication, is referred to as a smartphone. Preferably, the smartphone is equipped with a magnifying device, for example a magnifying lens, and / or a flash.

It is also possible to implement a so-called super-resolution technique obtained by sub-pixel assembly of images captured during a stream of candidate images by virtue of movements, possibly involuntary, of the user of the smartphone.

During an inspection phase, the package generally has cellophane wrapping, which can generate reflections from the flash of the smartphone. The barcode can also be placed close to the folds of the cellophane film and dust residues can also become embedded under the cellophane film, which generates artefacts or interference on the control image taken by the smartphone.

For example, a tobacco packet illustrated in FIG. 5 presents the following parasites, each of them being able to disturb the optical control: in reference 1, a fold of the cellophane packaging; in reference 2, reflections from the flash; in reference 3, a deformation of the points of the bar code after packaging; and in reference 4, an encrustation of dust residues.

* Footprint calculation

The footprint calculation during the control phase comprises the steps described below. A set of candidate images each comprising the bar code is obtained, for example in the form of a stream, that is to say of a sequence of video frames obtained by filming the bar code to be checked by the smartphone. Each frame can be a candidate image.

1. Localization of the barcode and correction of the candidate image

In this first step, the location of the bar code and its orientation in a 3-dimensional space are determined, for example using edge detectors and a frequency test.

The location of QR Codes is known, for example from “Fast QR Code Detection in Arbitrarily Acquired Images”, Luiz Belussi & al, August 2011, DOI: 10.1109 / SIBGRAPI.2011.16.

It is thus possible to determine the 4 extreme points of a rectangular barcode, therefore its orientation in space. This makes it possible to calculate a pose vector, used to rectify the control image.

The dimensions of the candidate image can be advantageously reduced by a predetermined factor, which makes it possible to speed up the processing.

A preprocessing is applied to the candidate image to eliminate the noise that can disturb the processing, for example by applying a Gaussian blur, as illustrated in FIG. 7.

To locate the barcode, initially we plan to apply a binarization method, in this case the Otsu binarization, as illustrated in figure 8.

On the binary image obtained, different morphological operations are applied to highlight the shape of the barcode (square, rectangular or other shapes), by a mathematical morphology operator, in this case to group the blobs, for example. at least one of dilation, erosion and the application of a morphological Laplacian. By blob is meant the object on the candidate image which is interpreted as corresponding to a module of the bar code.

In the example of the bar code illustrated in FIG. 9, a series of erosion followed by a closure is applied, thus making it possible to link the dots and reveal the square shape.

Then, a simple contour detection makes it possible to locate the bar code and thus deduce its orientation, as illustrated in figure 10.

The pose vector gives the orientation of the barcode but not its scale.

It is therefore possible to provide a rectification step (of angle and of scale) applied to the candidate image, using a 2D model which is specific to the shape of the bar code. In the example of Figure 8, the bar code tilts to the left. The applied 2D model consists of three points, corresponding respectively to the top-left, bottom-left and bottom-right points of the barcode, and makes it possible to straighten the candidate image of the barcode, as shown in the figure. 11.

2. Grid of the barcode and the exact location of the points constituting it

We apply a 2D grid calculated from the location of the edges of the bar code and the a priori knowledge of the number and dimensions of the points of the bar code. On this grid, provision is made to locate the locations of the points constituting the bar code, according to a predetermined model which depends on the shape and type of the bar code. Each cell in the grid corresponds to a point in the barcode. Each point includes a blob if the corresponding module of the square code is on (white) or does not include anything if the corresponding module of the square code is off (black).

An example is illustrated in FIG. 12 with a regular square grid for a datamatrix type barcode.

Before applying any algorithm, it is necessary to be able to locate oneself in the cluster of points, or blobs, of the candidate image. For this, it is necessary to be able to organize this set in the form of a grid. Filtering operations, for example with a frequency filter, are applied to reduce the noise and separation operations, typically mathematical morphology operators to separate the blobs, are optionally performed to divide the blobs which are stuck together. An erosion step can be implemented to remove the blobs of size less than a threshold value.

Then, the centroid of each blob is determined, by a known algorithm, in this case by a calculation of moments. A clustering operation is then provided on each vertical and horizontal coordinate of the set of blobs, in order to separate them according to one of the two criteria: row, horizontal, or column, vertical. This is possible because the number of columns and rows of blobs on the candidate image is known. Each cluster therefore corresponds to a row or a column depending on the dimension to which it is applied, see for example Figure 12.

It is thus possible to recover in a robust way the vertical and horizontal position of a blob and to determine its coordinates on the newly formed grid. Figure 13 illustrates the new grid generated with the coordinates of each blob. Note that spaces without blobs leave an effective void on the grid.

A rectified and cleaned candidate image is thus obtained.

3. Determination of the Unique identifier entered in the barcode

From this rectified and cleaned candidate image, a barcode decoding method can be applied as a function of the type (lD, 2D, QR code, PDF417, etc.) thereof. For example for a datamatrix, we can use the DMTX library. Once the identifier (UID) has been decoded, this code can be used to identify the fingerprint during subsequent processing and its transmission, for example to a database.

The UID thus known also makes it possible to know the reference bar code, that is to say that the position of the white or black modules is known. Thus, the positions at which blobs in the candidate image should appear are known.

It is then possible to calculate a set of descriptors characterizing each of the points of the candidate barcode. ²

In order to limit the influence of the possible parasites mentioned above, provision is made to implement at least one, and preferably at least two, of the following methods. The advantage of several methods is that they benefit from their complementarity. The proposed solution gains in robustness. For example, method 5.2 below is more sensitive to the problems associated with the deterioration of the points after cellophaning, while method 5.1 remains relatively robust to this type of deterioration.

4. Selection of barcode points

Once the 2D grid is in place, we can perform a cleaning step, in which we start by eliminating the positions of the grid that do not correspond to a laser engraving of the bar code, for example the empty spaces between the engravings, those where there is no blob, non-centered spots, lines or traces of cellophane wrapping the tobacco package. For example, if a white point on the candidate image corresponds to a position in the reference image for which the module is black (off), then this point is eliminated, i.e. unprocessed, because it should not match a laser engraving blob.

Then, at least one of the steps can be provided from:

- eliminate, that is to say do not process, the points corresponding to the identification elements specific to each bar code and which are invariant,

- select a predetermined area of the reference image, such that the selection corresponds to a sub-part of the barcode, for example by selecting an area of 10x10 modules instead of the usual 24x24 for a datamatrix, which makes it possible to speed up processing and reduce the size of the imprint. In this case, for the candidate image, provision is made to select an area with the same coordinates for the footprint calculation.

The term “identification elements” is understood to mean predetermined elements inherent to the type of bar code. For example (figure 8) a datamatrix comprises 4 marking elements in the form of lines on its edges: a continuous line at the bottom and another on the left, a black / white dotted line on the right and another on the top. In this case, the continuous registration lines (invariant in black) are eliminated. A QRcode consists of 3 marking elements in the form of landmarks. Etc. These marking elements make it possible to determine the scale of the barcode, and therefore the dimensions of its modules.

In some cases the choice of points to keep is directly linked to the method of calculating the footprint selected. It then takes place in step 5 below.

5. Calculation of Properties

It is planned to implement at least one, and preferably at least 2, of the following methods.

Each of these methods makes it possible to calculate a respective footprint. The footprints corresponding to each method are independent, but in the footprint storage formats, the “heavy” format described later contains a binarized summary of the image from which all of the footprints can be regenerated. It is also possible to store the fingerprints in vector form. In particular, it is possible to provide a vector with N components, with N the total number of fingerprint calculation methods, the position of each component in the vector corresponding to a predefined method.

5.1. Analysis of the area ratios of pairs of blobs

The appearance of the candidate image, or control image, taken with a windshield during the control phase is different from the reference image of the bar code taken during the enrollment phase. Thus, implementing a calculation of raw metric such as the area or the perimeter of each blob, and comparing the results between the reference images and the candidate images, can lead to results which are not satisfactory.

However, the variations which affected a given point of a 2D barcode, for example a datamatrix, whose candidate image was taken with a smartphone, normally affected another neighboring point in the same way. It can thus be interesting to calculate the ratios between these points.

To this end, image processing is provided for each reference image.

An original image is acquired in RGB color by a camera placed directly on the production line before packaging the package. The original image is then transformed into grayscale.

It can then be straightened, for example based on the 3 extreme points forming the square of a datamatrix, which amounts to applying a grid in which each cell is either empty or includes a point of the datamatrix, as illustrated in figure 14.

* Calculation of the area of each point of the datamatrix.

From the binary image (“binary” in the sense that each cell is empty or includes a point) calculated in the previous step, each point of the datamatrix forms an object from which the corresponding characteristics are extracted.

For example for the first 2 lines only of the datamatrix of figure 14, we obtain the table of figure 15 in which the indices indicate the position of a point in the datamatrix (on, white), and a dash indicates that it there is no point at this position (point off, black).

For all the points on, we extract the surface, that is to say the area. For example, after having binarized the image and dropped the grid, we can consider each blob as an independent connected object. Since it is a related binary object, white on a black background, it is possible to extract its area by counting the number of white pixels. For example for the 2 lines mentioned above, we obtain:

Area _(0.0) = 748, Area _(0.2) = 814, Area _(0.4) = 861, Area _(0.6) = 979, Area _{(0.8) =} p899, Area _(0.10 ) = 936, Area _(0.12 ) = 868, Area ₍₀₁₄₎ = 972, Area _{(016) =} 875, Area ₍₀₁₈₎ = 886, Area _{(0, 20) =} 896, Area _(0.22) = 903,

Area _(1.0) 1024, Area _(1.2) = 955, Area _((1.3) = 952, Area _(1.5 p899, Area _(1.9) = 1107, Area _((1.10) = 1041, Area _{(0, 13)} = 1213, Area _{(0, 14)} = 1048, Area _{(0, 16)} = 1002, Area _{(0, 20)} = 1094, Area _(0.22) = 1039, Area _{( 0.23)} = 981

By extracting the surface of all the points of the reference barcode, it is thus possible to construct a vector V_Ref whose elements are the surface of the lit points of the datamatrix according to a predetermined order, in this case from left to right and from top to bottom.

Let V_Ref l, m = Surface (i, j) with i = 0: 23; j = 0.23; m = 0: number of points lit For example, we obtain for the 2 aforementioned lines: V_Ref 1.0 = Area (0.0); V_Ref 1.1 = Area (0.2); V_Ref 1,2 = Area (0.4); V_Ref 1.3 = Area (0.6); V_Ref 1.4 = Area (0.8); etc. ; V_Ref 1.23 = Area (0.23).

We can then calculate surface ratios of each point with its neighbor, over the entire datamatrix. There are several possible implementations, for example following a path from left to right and from top to bottom, a point at the right end of a line being able to be neighboring with a point at the left end of the following line, as illustrated. by the arrows in figure 16.

The whole of the reference datamatrix is thus obtained in the form of a vector which contains all of the surface ratios of 2 successive points and which thus forms the imprint of said datamatrix.

We can also calculate V_Ref_Div i _{, m} = V_Ref i _{, m} / V_Ref i _{, m + i} with m = 0: number of points lit-1

Where V_Ref i _{, m + i} is the surface of the next element in the vector according to the chosen path, in this case from left to right and from top to bottom.

For the control phase, the same algorithm is implemented on the candidate image as on the reference image. We thus obtain:

V_Can l, m = Area (i, j) with i = 0: 23; j = 0.23; m = 0: number of points lit in this example; and we can calculate the vector which contains the area ratios between the points of the datamatrix by going from left to right and from top to bottom, namely:

V_Can_Div i _{, m} = V Susi _{, m} / V Sus i _{, m + i} with m = 0: number of points lit

With as V referring to a vector, the index Ref for the reference image, the index Can or the index Sus for the candidate image (also called suspect).

For the control phase, the reference datamatrix (s) which contain the same number of points, at the same position, are extracted from the database. The size of the 2 area ratio vectors, reference and candidate, which are datamatrix fingerprints, must be the same.

For each reference image, the absolute value of the difference between the vector of candidate ratios and the vector of reference ratios is calculated each time. If the difference of a pair is smaller than a predetermined value threshold, then an authenticity score of the datamatrix is increased, in this case a score which is initialized at 0 and incremented by +1 each time the condition is validated, of the candidate image. Then, we calculate the percentage of reports that are smaller than the threshold compared to the total number of reports established, which is the total size of the vector, namely:

V_Diff ⁿ _{1, m} = | V_Sus_Div _{1, m} -V_Ref_Div i _{, m} ⁿ | with n = l: number of reference images

In this case, the value of the threshold is between 0.04 and 0.12, and for example equal to 0.08. If the difference is strictly less than this threshold, we increase the authenticity score by 1. Then, we divide the score obtained by the size of the vector and we multiply by 100, if this percentage is strictly greater than a threshold value, by example between 60% and 99%; and in this case equal to 78%, the candidate image is authentic. The values can be chosen empirically from the database available.

Note that the route described here from left to right and from bottom to top can also be from right to left, from bottom to top, in column or diagonal, or non-linear. It just matters that it is predefined.

5.2 Angular Position of Blob Irregularities In this approach, the angular position of specific points located on the contour of each blob is determined. We choose two points, a first whose distance from the centroid of the blob is the smallest and a second point whose distance from this same centroid is the highest. For each point, the angular position in a predetermined coordinate system is calculated.

We can then, by superposition of the centroids, compare the angular positions of the specific points for a blob of the candidate image with the angular positions of the specific points for the corresponding blob of the reference image, and consider that the 2 blobs are identical if the values of the angular positions are less than a predetermined threshold value, and that the 2 blobs are different otherwise.

For example in figure 17 and figure 18, we have a blob of the candidate image CAND of centroid O ', a first point 10' whose distance from the centroid O 'of the blob is the greatest and a second point 20 'whose distance from this same centroid is the smallest. We also have a blob of the reference image REF of centroid O, a first point 10 whose distance from the centroid O of the blob is the greatest and a second point 20 whose distance from this same centroid is the smallest. . The 2 blobs are superimposed by their respective centroid.

In figure 17, the angular positions of the specific points between the CAND blog and the REF blob are the same, for example by calculating the angle between the line O'-10 'and the line O'-20' and by calculating the 'angle between line 0-10 and line 0-20. These two blobs are then considered the same.

In figure 18, the angular positions of the specific points between the CAND blog and the REF blob are different, for example by calculating the angle between the line O'-10 'and the line O'-20' and by calculating the angle between line 0-10 and line 0-20. These two blobs are considered different.

We can calculate a similarity score between two sets of blobs.

We can also ignore the blobs which present an irregularity of contour which is not marked enough, for example by making the difference in distance between the point of minimum distance 20, 20 'and the point of maximum distance 10, 10' and d 'ignore this blob if this difference is too small, that is to say less than a predetermined threshold value.

5.3 Mapping of Blob Irregularities

This approach involves spotting the main irregularities in a blob and writing them down on a map so that you can compare them.

On each blob, we recover its contour and we determine, with the gradients, the orientation of this same contour. We can represent the different orientation values with colors. Once this orientation map has been obtained for a blob, we try to compare it to a perfect orientation map, which corresponds to the orientation map of a circle.

Thus, at each point of the contour, we compare its orientation with that of a perfect circle. If the deviation of the orientations exceeds a predefined threshold value, then this point of the contour is marked on a new map, which is the map of the irregularities of the processed blob.

Thus, compared to the map of the orientations of a circle, it is possible to distinguish points where the hollows or bumps are too large compared to the threshold.

We proceed in this way on each blob to obtain a multitude of irregularity cards, specific to each one. In a comparison of two blobs, it can be determined whether the irregularities that are common and located in the same places are large enough to decide whether the blobs are similar and to obtain a similarity score between each blob. 5.4 Building a blob profile

This approach consists in constructing a curve characterizing the profile of a blob. It then makes it possible to carry out a comparison of profiles between the profile of a blob of the candidate image and the profile of the corresponding blob of the reference image.

The profile of the blob is determined by extracting the distance to the centroid of each point of the outline of the blob, along a predetermined path, preferably over the entire outline of the blob.

A profile is thus obtained for a blob of the candidate image and a profile for the corresponding blob of the reference image. From these profiles, one or more metrics can be extracted in order to determine whether these two profiles are similar or not. For example, we can measure the correlation between the two profiles to determine if they are similar or not.

For example in Figure 19, the two profiles illustrated have a correlation of 0.77 and the blobs are considered identical.

In Figure 20, the two profiles shown show a correlation of -0.07 and the blobs are considered to be different.

5.5 Characterizations based on the oriented gradient histogram (HOG)

Each point extracted from the binary image is characterized by a set of histograms, calculated from the intensity and orientation of the gradients. Starting from the center of each point (or blob), its bounding box is centered in a 64x64 pixel window which is segmented into 8x8 blocks. Starting from these 8x8 blocks, the histograms are calculated from the intensity and the orientation of the gradients of each pixel constituting each of the images (8x8 pixels). The histograms are calculated on blocks of 8 × 8 pixels, then normalized by blocks of 16 × 16 pixels, by concatenating the 4 histograms. The 16x16 blocks overlap by 8 pixels. Starting from the centroid of each point, the binary image of the latter is centered in a window of 64x64 pixels. This size should actually be a multiple of 8 and may vary depending on the grinding pattern and stitch size. For the example described here, the size of the bounding box of the points is limited to 64x64 pixels. Then we calculate the normalized histograms on this image. Thus each point is represented by 6x6 histograms allowing the local characterization of its shape.

In the general case, the descriptor of an image I of a 2D code represented by a grid of NxM consists of all the descriptors of all the points detected.

Thus, we have: Descri ptor (l) = {descriptor (pi _j ), iε [l..N], jε [1..M]} With p _i , j a detected point

The descriptor of a point p at position i, j in turn consists of the set of histograms of the oriented gradients (Hog) of all the blocks of size B × B pixels which overlap by B / 2 pixels. The total number of blocks extracted from the image of point p is K x L.

We thus have: descriptor (p _ij ) = {Hlog (block _{k, i} ), kε [l..K], lε [l..L]}; with K = (w /0.5*B)-1 and L = (h / 0.5 * B) -1; where h and w represent the dimensions of G image of point py

The measurement of similarity between the reference image and the candidate image consists firstly in calculating a similarity score for each pair of “reference-candidate” points. Then, all of these scores are used to calculate an overall score measuring the similarity between the two images.

It is possible to carry out a measurement of the local similarity, by pairs of points.

Based on the 2D grid of each of the two candidate and reference images, the matching of the points according to their position in the grid makes it possible to constitute pairs of points. A homogenization of the size of the points (reference, candidates) is necessary for each pair in order to overcome the problem of variation of the acquisition conditions between the enrollment system and the control system.

Thus, the bounding box of the candidate point is resized to the size of the bounding box of the corresponding reference point.

The similarity between a reference point p_Ref and candidate point p_Cand is measured by the average of the distances of the histograms calculated from the images of the bounding boxes of each of the two points. For this purpose, we can use the Euclidean distance, or other distances.

We thus have:

Similarity (p_Ref, p_Cand) = Σ ^KxL i = i distance (Hog _Ref [block i], Hog cand [block i]) / KxL

With the image of each of the bounding boxes made up of KxL blocks.

As an alternative or in addition, a measurement of the overall similarity can be carried out.

In this case, the measurement of overall similarity between the reference image and the candidate image consists in calculating the rate of reference points whose distance is less than a predetermined threshold.

Another measurement can also be carried out, by assigning a greater weight to the most relevant reference points, i.e. which present a deformation linked to the hazard of the laser engraving and which turn out to be richer in terms of reference. 'a point of view of their characterization, that is to say whose contour is not perfectly circular.

5.6 Local dissimilarity map (CDL)

A local dissimilarity map (CDL) filtering step can be provided, possibly preceded by a denoising step.

A local dissimilarity map (CDL) makes it possible to measure the local differences between two binary images, ie two images in black and white (not in gray levels).

The calculation of a local dissimilarity map is a modified version of the Hausdorff distance, it is notably described by Morain-Nicolier, Frédéric & LANDRE, Jérome & Ruan, Su. (2009). “Object detection by measuring local dissimilarities. ".

We define by CDLbin the CDL of two binary images A and B as being:

C D Lb ï n (A, B) (p) = | A (p) -B (p) | max [dA (p), d B (p)] with p = (x; y) and dX (p) the transform into distance from image X to point p, where in this case X = A or B.

The application of a local dissimilarity map is known in particular in the medical field for the search for tumors, as described for example in the article “Localization of tumors in PET sequences, by detection of changes by means of local dissimilarities” Ketata et al. (CORESA 2013 - 16th edition of the compression and REpresentation of Audiovisual Signals conference Le Creusot, FRANCE, November 28 and 29, 2013).

Provision is made here to compare the differences between a first binary image A, for example a reference image, and a second binary image B, for example a candidate image.

It is also possible to provide for comparing the differences between a first sub-image of binary image A, and a second sub-image of binary image B, the position and dimensions of which are the same (at an enlargement ratio close). A first sub-image is first obtained by applying a digital mask comprising a first window F1 corresponding to all or part of the candidate image, which makes it possible to extract the part of the candidate image recorded in the first window Fl.

Similarly, a second sub-image is then obtained by applying a digital mask comprising a second window F2 corresponding to all or part of the reference image, which makes it possible to extract the part of the reference image entered in the second window F2.

The shape of the window F1 is identical to that of the window F2. The size of the window F2 is equal to that of the window F1, up to the scale factor.

If the candidate image is an already binary image, then the first binary image used for the calculation of the local dissimilarity map is the first sub-image obtained by the first window F1, and the second binary image used for the calculation of the local dissimilarity map is the second sub-image obtained by the second window F2.

CDL (F1, F2) is then calculated the CDL between the first sub-image obtained by the first window F1, and the second sub-image obtained by the second window F2.

Then we compare CDL (F1, F2) to a predetermined D_CDL threshold and:

- if CDL (F1, F2)> D_CDL then the candidate image is considered to be different from the reference image,

- if CDL (F1, F2) <D_CDL then the candidate image is considered identical to the reference image.

The graphic enhancement can be carried out by at least among: a modification of the value of the pixels (color, luminance), the display of a frame around the points or the display of a frame at the periphery of the windows Fl and F2, for example dotted or colored, etc.

If the candidate image is in gray levels, it is split into several binary images.

For example, the maximum value and the minimum value of the pixels of the candidate image are stored in a memory. The difference between the maximum value and the minimum value is divided into several intermediate threshold values, preferably at regular intervals.

For a first intermediate threshold value, all the pixels whose value is less than the value of said intermediate threshold are replaced by black and all the pixels whose value is greater than the value of said intermediate threshold are replaced by white. A first binary image is thus obtained. Then, the threshold value is varied and a second binary image is obtained with a second intermediate threshold value. And so on for the set of intermediate threshold values.

N binary images are thus obtained, with N a natural integer equal to the number of selected threshold values, ie the number of cuts.

For each binary image i (i between 1 and N), CDLJ (F1, F2) is calculated the CDL between the first sub-image obtained by the first window F1, and the second sub-image obtained by the second window F2.

We can then calculate the weighted CDL of the set of N binary images, that is to say that we calculate CDL (F1, F2) = 1 / N Σ CDL_N (F1, F2).

More generally, we can calculate the global CDL of the set of N binary images as being:

With Ai (respectively Bi) a binary cut i of A (respectively B), i between 1 and N.

CDL (F1, F2) is then compared to a predetermined D_CDL threshold, as described above.

If the candidate image is in color, it can be saved in grayscale and processed as described above.

Alternatively, provision can be made to convert the candidate image into a 3-dimensional or 3-channel color space, in this case in the RGB space or in the CIE XYZ space.

The candidate image is split into C binary images, with C a natural integer corresponding to the number of channels, in this case equal to 3.

Each of the C binary images can then be processed in a manner similar to the N binary images obtained when the candidate image is in gray levels.

Thus for each binary image C, CDL_C (F1, F2) is calculated the CDL between the first sub-image obtained by the first window F1, and the second sub-image obtained by the second window F2.

We can then calculate the weighted CDL of the set of C binary images, i.e. we calculate CDL (F1, F2) =

More generally, we can calculate the global CDL of the set of C binary images as being: CDL_XYZ (A, B) (p) = ±

With in this case C = 3

For two binary images we have:

C D Lb ï n (A, B) (p) = | A (p) -B (p) | max [dA (p), d B (p)]

For two grayscale images we have:

CDL (A, B) (p)

With N natural integer at most equal to the number of gray levels, and in this case equal to the number of binary images resulting from the processing of the original image in gray levels.

For two images with C channels we have:

CDL_C (A, B) (p)

With C the number of channels, and Ak the channel k e (X; Y; Z) of image A.

5.7 Detection of convexity defects

This method aims to detect points where the convexity is not perfect on the contours of each point of the datamatrix.

For a reference image, acquired by a camera placed directly on the production line before packaging, provision is made to record the reference image in the form of a binary image comprising the coordinates of the centers of each point of the datamatrix. Then we plan to calculate the contours of the points of the datamatrix from the binary image, that is to say to produce a curve joining the continuous points along the border of the pixels having the same intensity, in order to allow a form analysis.

We can then calculate the convex envelope of the point corresponding to a modulus of the datamarix, ie its contour.

Then, the convexity defects are calculated for each convex envelope. To determine the points of convexity defects, it suffices to compare the convex envelope of a blob with respect to its true contour. As long as there is a point of the true contour which comes out from one side or the other of the convex envelope, we can say that here there is a defect. Any deviation of the envelope from a perfectly convex envelope is considered a defect in convexity.

For a candidate image, the same algorithm is applied and the convexity defects of each blob are calculated.

It is then possible to calculate the distance between a defect of convexity of a blob of the candidate image and a defect of convexity of a corresponding blob of the reference image. If the distance is less than a predefined threshold value, then the value of a counter is increased, for all the blobs of the datamatrix.

Once all the blobs have been processed, we can compare the counter value to a predefined threshold value and if the counter is greater than the threshold value then the candidate image is considered similar to the reference image, so the candidate barcode is considered identical to the reference barcode.

It is also possible to calculate the percentage of the number of points found on the candidate horn image exhibiting a convexity defect relative to the total number of convexity defect points on the reference image. From a certain threshold percentage, the candidate datamatrix is considered authentic.

5.8 Calculation of moments

A reference image is generally an RGB color image, acquired by a camera placed directly on the production line before packaging.

Provision is preferably made to readjust the reference image, for example using 3 end points forming the square of the datamatrix as explained previously.

It is planned to transform the reference image into gray levels.

Then, provision is made to calculate the grid, as explained previously, which produces a binary image with the coordinates of the centers of each point of the datamatrix.

It is then possible to calculate the external contours of the points of the datamatrix from the binary image, as explained above.

For each contour, we can then calculate the following different moments.

Spatial moments: mOO, mOl, mlO for 3 cases 00, 01, 10.

Or more generally: m _ji =

With array the function array.

Centered moments: m20, mil, m02, m30, m21, ml2, m03 for cases 20, 11, 02, 30, 21, 03, 12 Or more generally: (x - x_) ^j (y - y_) ⁱ )

With x_ety_ the centers of mass such as x_ = ml0 / m00 ety_ = m01 / m00 Normalized centered moments: nu20, null, nu02, nu30, nu21, nul2, nu03

^M1 "ί

For each blob, we can have a vector V_Ref ^m with m the index of the blob.

Let V_Ref ^m = [m00 ^m , m ₀₀ ^m , m ₀₁ ^m , m ₂₀ ^m , m ₁₁ ^m , m ₀₂ ^m , m ₃₀ ^m , m ₂₁ ^m , m ₁₂ ^m , m ₀₃ ^m , nU ₂₀ ^m , nu ₁₁ ^m , nude ₀₂ ^m , nude30 ^m , nude ₀₂ ^m , nude ₀₂ ^m , nude ₀₃ ^m ]

With m = 0: number of points lit

We can thus calculate the set of vectors V_Ref ^m for the set of m blobs. All of these vectors constitute the imprint of the datamatrix of the reference image.

For a candidate image taken with a smartphone, the same algorithm is applied to calculate the value of the vector V_Cand ^m .

Since the coordinates of all the points which form the grid are known, the points of the reference image can be matched with those of the candidate image. In fact, for each point of the datamatrix of the candidate image, we look for the corresponding point in the reference image having the closest coordinates. Thus, we have pairs of points.

The comparison is made by metric between the moments, then by a rule on the sets of points.

For each pair of points, we can then calculate the differences between the moments. V_Diff ⁿ = | V_Cand ⁿ -V_Ref ⁿ | with n = l: number of reference images.

Preferably, the normalized centered moments are chosen, and more precisely the nu20 and null moments.

When the difference between the moments for a given pair of points is less than a predetermined threshold value, a score of the candidate datamatrix is incremented.

We can then calculate the percentage of point pairs having small differences in moments from the total number of matched point pairs. If the percentage is greater than a predefined threshold, or when the score is greater than a predefined threshold value, then the candidate datamatrix is considered authentic.

The invention is not limited to the times described above. We can foresee other moments like Zernike, Hu, Legendre, etc. then to build an imprint from those moments.

5.9 Topological Analysis of Irregular Points (TDA)

Topological Data Analysis (TDA) is a mathematical method that extracts information from a data structure. With the recent development of persistent homology, it is possible to efficiently infer the geometric structure of an object if the geometric structures of a pair of points are similar, for example as described at http: // outlace .com / TDApartl.html.

5.9. To. Similarity measure between two points In a first variant, provision is made to compare the topological structure of a reference blob and of a candidate blob.

After filtering the blobs of the datamatrix which are of no interest other than reading the encoded UID (for example a blob without defect), provision is made to select a blob to be analyzed.

After binarization, we transform the contours of the candidate blob into a cloud of points.

According to a first variant for constructing this cloud of points, the centroid of the selected blob is calculated, and a finite number of points is projected onto its contours along the axis of the centroid. The illustration of this method corresponds to FIG. 21 lower right. First, we calculate the centroid of the blob, then we project on the contour of the blob a finite number of points. We can imagine a trigonometric circle whose origin of the circle is the centroid of the blob. We turn around the circle with a constant angle and we place a point at each incrementation of displacement.

A second variant of construction of the cloud of points consists in traversing the contour of the selected blob in a geodetic manner and in drawing a point every n lengths until having traversed the entire contour of the blob. This method is illustrated at the top right of FIG. 21. We start by calculating the total distance from the outline of the blob. Then, we choose a number of points to draw on the outline. If we want to place n points, we must divide the perimeter of the blob by n.

A third variant for converting the contours into a cloud of points consists in using an approach similar to that which is described for the Angular localization of the Irregularities of the Blob.

These 3 variants are illustrated in FIG. 21. In this figure, on the left is illustrated a binarized blob.

In the center is illustrated the outline of said blob. The result of the first variant is illustrated at the top right; that of the second variant is illustrated in the center on the right, and that of the third variant is illustrated at the bottom right.

From the point cloud, we plan to make geometric constructions, called filtrations. Filtration, in topological analysis, makes it possible to bring out a topological structure of a set of data. The objective is to build families of objects at different scales and to study their evolution. The result obtained after filtration is then represented in the form of a persistence diagram using the information tools of persistent homology, the information tools being the diagrams of the domain of persistent homology shown in FIG. 22 and FIG. 23, that is to say the representation of the change of scale of the points, the bar code of persistence and the persistence diagram.

According to the invention, provision is made to construct a persistence diagram from the point cloud. During the filtration step, the diameter of the points is gradually increased and the radius of each point is reported in the form of a persistence bar code.

Each time an intersection appears between two points, one considers that one of the points dies and one stops following its evolution on the persistence barcode. The resulting barcode is converted into a persistence diagram. We then obtain a topological descriptor which can be used in a training model. We stop making the balls grow once a single component evolves on the persistence diagram. Depending on the arrangement of the initial points, cycles may appear during the filtration stage. When a cycle appears, its evolution is followed on the persistence barcode. Filtration is finished when there is only one point left which evolves (see Figure 22). In fact, the points are increased incrementally (on the left in FIG. 22) and at the same time we follow the evolution of the radius of each point by building a persistence bar code (on the right in FIG. 22). A bar (top right FIG. 22) corresponds to the evolution of the radius of a point. The two isolated bars under the barcode (center right and bottom right FIG. 22) correspond to the appearance of a cycle. A cycle appears when the "balls" touch each other. A cycle dies when the center of the cycle is "clogged". A cycle is illustrated in the second image of FIG. 22. On the x-axis of persistence barcode, we have the lifespan of the "balls" / cycles and on the ordinate "the identifier" of the point / cycle.

In order to be able to obtain a topological descriptor from the persistence bar code obtained, provision is made to implement a persistent homology method.

Finally, we use the topological descriptor of the persistence diagram, which is passed into a classification model (classic Machine Learning or neural network), to determine the similarity between two blobs.

5.9. b. Topological analysis of the set of irregular points

In a second variant, the same principle is applied to all of the blobs which have defects rather than to the point cloud of the contours of a single blob.

5.9.c. Topological graph

In a third variant, provision is made to construct a graph from the blobs which have defects. We can add an edge between two neighboring points whose contours include irregularities.

6. Storage of the impression

At least one of the following two storage modes can be provided.

6.1. A "heavy" mode, the size of which is in this case about 20 kb. In this mode, after preprocessing and rectification according to the 2D model of the grid, the image of the 2D code is stored in a compressed binary format (for example .png).

6.2. A "light" mode, the size of which is in this case about 5kb. In this mode, we store the coordinates of the center of the points in the 2D model, as well as all the metrics associated with them. The footprint is smaller but it is not scalable, unlike the "heavy" mode.

7 - Preparation for an Optimal Comparison

A preparation step can be provided for an optimal comparison.

This step can be implemented upstream of the calculation of the footprint. In this case the imprint can be stored in a particularly compact form. It can also be implemented downstream from the calculation of the footprint, for a greater richness of the representation at the time of the comparison. For this purpose, the data conveyed by each of the fingerprints (corresponding to each of the calculation methods) is considered as a set of signals. These signals in turn characterize the link that exists between the blobs. This information can be quantified before the footprint is stored, which makes it possible to reduce the size of the overall footprint and thus optimize the comparison. It is also possible to store the fingerprints calculated from each of the methods (larger storage) and at the time of the comparison, the quantification is applied. In the first case, we benefit from compact storage but with a non-scalable quantification method. In the second case, the raw information of the fingerprints is stored, but flexibility is gained because the quantification and the comparison process can be continuously improved, ie combining different quantification methods, different comparison heuristics.

In this step, the information corresponding to each of the fingerprint calculation methods, that is to say the signals conveyed by the fingerprints, are reduced to a quantified representation space, such as for example a binarization, by following a process which preserves the most important characteristics of these signals, i.e. the variation of fingerprints from one blob to another and the link that exists between neighboring blobs, while allowing a point-to-point binary comparison induced values, which depend on the quantification method. In the case of binary quantization, these are the binary sequences. Thus, the quantization of the signals conveyed by the imprints makes it possible to have a reduced and homogeneous representation space and consequently an optimal comparison process. This avoids adapting the process to each of the methods (with its own representation space, and its own similarity measures). For example, with binary quantization, the comparison process consists in comparing the binary sequences and thus benefiting from the comparison methods which are known for their speed.

One method of doing this may be the representation of the fingerprints as one or more bit sequences. Thus the comparison between reference fingerprints and candidate fingerprints amounts to comparing the sequences of bits which represent them. This comparison can be performed, for example, by a simple count of difference bits. Other measures adapted to binary sequences can be envisaged, for example the Hamming distance, the rate of error bits per bit blocks, etc.

A sequence of bits is calculated from the variation of the fingerprints between two neighboring blobs by measuring the sign of the differences of the corresponding fingerprints. The sequence of bits between two blobs i, j consists of the concatenation of the binary representations extracted from the fingerprints of each of the methods. For example, for a method of analyzing the area ratios of pairs of blobs (see paragraph 5.1 of the present description), the binary sequence is extracted from the sign of the difference between the areas of the two neighboring blobs i and j.

If we denote by k the number of methods chosen, the sequence of bits characterizing the link between two neighboring blobs is defined as follows:

That is to say that the calculation of the sequence of bits is adapted to each of the fingerprint calculation methods chosen.

For example, if we take the first three methods described here.

For the method described in 5.1, the sequence of bits for the two neighboring blobs i, j can be calculated as follows:

For the method described in 5.2, the bit sequence for the two neighboring blobs i, j can be calculated as follows: with RE {1,2} corresponding to the two irregularities detected by method 2

For the method described in 5.3, the bit sequence for the two neighboring blobs i, j can be calculated as follows: Thus a 2D bar code is represented by the set of bit sequences of all the pairs of neighboring blobs which constitute it. The type of neighborhood is defined beforehand.

Barcode decoding

For the decoding of the datamatrix in order to obtain its UID, which serves as an index in the database, a step of cleaning the candidate image can be provided, in this case by removing the parasites and spots whose surface is greater than that of a data matrix module, for example by applying a frequency filter. It is also possible to provide a step of detecting the quality of the candidate image, in this case by virtue of image quality detectors (or analyzers). Depending on the response of the quality detectors, that is to say if the desired quality is not reached, the current candidate image is dropped and a next candidate image of the stream is selected. An attempt is then made to decode the bar code present on the new candidate image. If the code is not readable or does not correspond to a (known) rule for constructing the unique UID number, then we go to the next candidate image of the stream.

Any cleaning is only used for decoding the bar code, which makes it possible, via reading the UID, to search for the corresponding fingerprint in a database. The processing steps described above are performed on the straightened (uncleaned) candidate image.

Thanks to the invention, it is possible to respond simultaneously to the following four strong constraints

- Obtain a very low rate of false negatives, corresponding to authentic packets considered to be suspect,

- Obtain an extremely low rate of false positives, corresponding to false packets considered to be true,

- Have a footprint of the smallest possible size, for the storage and transport of the package,

- Calculate this footprint as quickly as possible (less than the production time of the package).

Finally, the comparison time between two fingerprints does not necessarily have strong constraints in terms of calculation time.

Claims

1. A method of enrolling a reference two-dimensional bar code, by processing a reference image of said reference two-dimensional bar code, the bar code comprising a set of dots having random printing defects, each point of the reference barcode corresponding to a respective blob in the reference image, the method comprising the steps of:

- save in a memory a reference image in gray levels of the reference barcode,

- recording an identifier (UID) of the reference barcode in the form of an index in a database, characterized in that it further comprises steps consisting in:

- calculate a reference imprint of said reference bar code, the imprint being a function of the random printing defects, and

2. The enrollment method according to claim 1, wherein when the reference fingerprint in the database is recorded in the form of a vector, said vector of the reference fingerprint comprises several components, at least one of which is non-zero, and each component is calculated by one of the respective steps among:

- determine a map of the irregularities of the blobs;

3. A method of verifying a candidate bar code comprising an identifier (UID), the method comprising the steps of:

- record in a memory a candidate image in gray levels of the candidate barcode,

- decode the identifier (UID) of the candidate barcode,

- identify the reference bar code according to any one of claims 1 or 2, of which the index in the database is equal to said identifier of the candidate barcode,

- calculate an imprint of said candidate barcode,

4. The method of claim 3, wherein the step of comparing comprises at least one of the steps among:

- a topological analysis of irregular points.

5. Method according to any one of claims 3 or 4, comprising the steps of:

- determine the location of the candidate barcode and its orientation in a 3-dimensional space,

- calculate a pose vector,

- straighten the image of the candidate barcode, and

- optionally clean said image of the candidate barcode.

6. Method according to any one of claims 3 to 5, comprising a step of representing the imprint of the candidate bar code and the imprint of the reference bar code by one or more sequences of bits,

7. Method according to any one of claims 3 to 6, in which the reference bar code and the candidate bar code are two-dimensional bar codes, the method further comprising at least one of the steps among:

- apply a frequency filter to the candidate image,

- detecting the quality of the candidate image, using image quality detectors.

8. A method according to any one of claims 3 to 7, wherein the reference bar code and the candidate bar code are marked on a tobacco package, optionally covered with cellophane.

9. A computer program comprising program code instructions for executing the steps of the method according to any one of claims 3 to 8, when said program is executed on a computer.

10. Smartphone comprising a memory in which is stored the computer program according to claim 9.