Method and device for lossy compression relative to areas of interest of digital images using parameters determined from a neural network.
The neural network-based lossy compression method optimizes bit allocation in areas of interest to reduce distortion and maintain detection accuracy in reconstructed images, addressing detection errors caused by existing compression techniques.
Patent Information
- Application Number
- FR2023004591
- Authority / Receiving Office
- FR · FR
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-09
- Publication Date
- 2024-11-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing lossy image compression methods cause detection errors in reconstructed images due to degradation, affecting the accuracy of object detection in areas of interest.
A lossy compression method using a neural network to determine coding parameters through a learning phase that minimizes distortion in areas of interest while optimizing bit allocation, employing gradient descent to adjust neural network parameters based on pixel weighting coefficients.
The method reduces detection errors in reconstructed images by concentrating bits in areas of interest, maintaining high visual quality and detection accuracy while minimizing overall distortion.
Smart Images

Figure 00000000_0000_ABST
Abstract
Description
Title of the invention: Method and device for lossy compression relating to areas of interest in digital images using parameters determined from a network of neurons. Technical field
[0001] The present invention relates to lossy compression of digital images and the detection of objects in said digital images. More particularly, the present invention relates to a method and device for lossy compression and relating to the areas of interest of a digital image using parameters determined from a neural network. Technological background
[0002] It is known to compress (encode) lossily digital images to minimize the memory footprint of these digital images when these digital images are stored or to minimize the quantity of bits, i.e. the flow rate, necessary to transmit these digital images between two remote devices via a communication network.
[0003] There are many lossy image compression methods. Some of these methods use neural networks to determine coding parameters used by these compression methods.
[0004] Furthermore, it is common to apply a detection method to a reconstructed digital image obtained by decoding a compressed digital image to detect objects present in this reconstructed digital image. The detection method determines areas of the reconstructed digital image which carry semantic content of this reconstructed digital image. These areas are often called areas of interest or objects of the reconstructed digital image.
[0005] When a detection method is applied to a reconstructed digital image, detection errors may occur that do not exist when this detection method is applied to the original digital image. These detection errors are due to degradations of the reconstructed digital image caused by lossy compression of the original digital image. Summary of the invention
[0006] An object of the present invention is to provide a lossy compression method for a digital image which minimally degrades the performance of a detection method applied to a reconstructed digital image.
[0007] According to a first aspect, the present invention relates to a method for lossy compression of a digital image using coding parameters determined from a neural network, said method comprising a learning phase for determining values of the parameters of the neural network which comprises the following steps for a set of current digital images: a) obtaining a weighting coefficient value for each pixel of each current digital image of said set of current digital images, said weighting coefficient value depending on whether the pixel of a current digital image belongs to a defined area of interest in said current digital image; b) selection of current values of the neural network parameters; c) obtaining a value of a cost function for each current digital image of the set of current digital images, said cost function being defined by a sum of a first data item and a second data item, said first and second data items being evaluated for said current values of said parameters of the neural network, the first data being representative of an average number of bits per pixel necessary to encode said current digital image using the current values of said parameters of the neural network and the second data being representative of a distortion between said current digital image and a digital image reconstructed from the current values of said parameters of the neural network, said second data being dependent on the values of the weighting coefficients; d) calculation of an average value of the obtained cost function values; e) determining step values according to a gradient descent method used to minimize the average value of the cost functions, each step value being used to modify a current value of a parameter of the neural network; steps c), d) and e) are iterated until a stopping condition of the gradient descent method is verified.
[0008] The learning phase makes it possible to determine values of the parameters of the neural network by a gradient descent method used to minimize an average value of cost functions calculated for current digital images of a set of digital images. The neural network is then trained and can be used in a production phase during which a digital image to be compressed is encoded according to its areas of interest from coding parameters obtained from the trained neural network.
[0009] According to a variant, the second data is a mean square error which measures the distortion between a current digital image and a re- digital image constructed corresponding to the current digital image.
[0010] According to a variant, the value of a weighting coefficient corresponding to a pixel belonging to an area of interest of a current digital image is greater than a value of a weighting coefficient corresponding to a pixel not belonging to an area of interest of said current digital image.
[0011] According to a variant, several areas of interest being defined in a current digital image, values of weighting coefficients corresponding to pixels, called first pixels, belonging to said areas of interest of said current digital image are fixed so that a distortion calculated between the first pixels of said current digital image and the first pixels of said reconstructed digital image does not exceed a threshold value and the values of weighting coefficients corresponding to pixels, called second pixels, not belonging to said areas of interest of said current digital image are fixed so as to distribute compression errors over the second pixels.
[0012] According to a variant, values of weighting coefficients corresponding to pixels belonging to an area of interest of said current digital image are fixed so that a distortion calculated between the pixels of said area of interest of said current digital image and the pixels of said area of interest of said reconstructed digital image does not exceed a threshold value.
[0013] According to a variant, the values of weighting coefficients are matched with a segmentation map representing areas of interest in a current digital image, said segmentation map can be obtained by a method of segmenting said current digital image.
[0014] According to a second aspect, the present invention relates to a device for lossy compression of a digital image comprising means for implementing steps of the method according to the first aspect of the present invention.
[0015] According to a third aspect, the present invention relates to a vehicle comprising at least one sensor and the preceding device.
[0016] According to a fourth aspect, the present invention relates to a computer program product comprising instructions adapted for executing the steps of the methods according to the first aspect when the computer program is executed by at least one processor.
[0017] Such a computer program may use any programming language, and be in the form of source code, object code, or intermediate code between source code and object code, such as in a partially compiled form, or in any other desirable form.
[0018] According to a fifth aspect, the present invention relates to a computer-readable recording medium on which a program is recorded. computer system comprising instructions for carrying out the steps of the method according to the first aspect of the present invention.
[0019] On the one hand, the recording medium may be any entity or device capable of storing the program. For example, the medium may comprise a storage means, such as a ROM memory, a CD-ROM or a microelectronic circuit type ROM memory, or a magnetic recording means or a hard disk.
[0020] Furthermore, this recording medium may also be a transmissible medium such as an electrical or optical signal, such a signal being able to be conveyed via an electrical or optical cable, by conventional or hertzian radio or by self-directed laser beam or by other means. The computer program according to the present invention may in particular be downloaded from an Internet-type network.
[0021] Alternatively, the recording medium may be an integrated circuit in which the computer program is incorporated, the integrated circuit being adapted to perform or to be used in performing the method in question. Brief description of the figures
[0022] Other characteristics and advantages of the present invention will emerge from the description of the non-limiting embodiments of the present invention below, with reference to the appended figures 1 to 3, in which:
[0023] [Fig-1] schematically illustrates the lossy compression process of a digital image using coding parameters determined from a neural network according to a particular and non-limiting example of the present invention.
[0024] [Fig.2] illustrates a flowchart of the different steps of a lossy compression method for a digital image using coding parameters determined from a neural network according to a particular and non-limiting exemplary embodiment of the present invention;
[0025] [Fig.3] schematically illustrates a device for lossy compression of a digital image using parameters determined from a neural network according to a particular and non-limiting exemplary embodiment of the present invention. Description of the embodiments
[0026] Methods and devices will now be described in the following with joint reference to Figures 1 to 3.
[0027] The lossy compression process comprises a learning phase during which the neural network is trained and a production phase during which the original digital image x is compressed from coding parameters determined from the trained neural network.
[0028] The learning phase comprises a series of operations applied to a set of N current digital images xn (ne { 1 ; TV] ) coming for example from the image bank BDD100K associated with an annotation file which contains information defining at least one area of interest for at least one current digital image xn.
[0029] In a first operation, a weighting coefficient value is obtained for each pixel ( / , j ) of each current digital image xn of the set of current images, the weighting coefficient value depending on the membership of the pixel ( i, J) of a current digital image x» to a zone of interest defined in said current digital image xn.
[0030] According to a variant, the definition of an area of interest of a current digital image is obtained from the annotation file associated with the image bank BDD100K.
[0031] In a second operation, initial current values of the parameters of the neural network are selected.
[0032] According to a variant, the initial current values are selected randomly.
[0033] In a third operation, a value of a cost function L is obtained for each current digital image of the set of current images. Said cost function L is defined by a sum of a first data item R and a second data item D.
[0034] The first data R is representative of an average number of bits per pixel necessary to encode said current digital image x» using the current values of said parameters of the neural network and the second data D is representative of a distortion between said current digital image x» and a reconstructed digital image xn from the current values of said parameters of the neural network. The second data D is dependent on the values of the weighting coefficients
[0035] In a fourth operation, an average value of the obtained cost functions L is calculated.
[0036] In a fifth operation, step values are determined according to a gradient descent method used to minimize the average value of the cost functions L, each step value being used to modify a current value of a parameter of the neural network.
[0037] The third, fourth and fifth operations are iterated until a stopping condition of the gradient descent method is verified.
[0038] The gradient method for determining parameters by minimizing a cost function is very widely known to those skilled in the art. One can refer for example to the following link for more details: https: / / moodle.luniversitenumerique.fr / pluginfile.php / 585 / mod_resource / content / l / S2 b-descente.pdf.
[0039] According to a variant, a stopping condition of the gradient descent method is a predetermined number of iterations.
[0040] According to a variant, after each iteration of the gradient descent method, the step values are compared to threshold values and if all these step values are less than the threshold values, a stopping condition of the gradient descent method is verified.
[0041] According to a variant, the second data D is a mean square error which measures the distortion between a current digital image xn and a reconstructed image xn corresponding to the current digital image.
[0042] For example, in the case where the current digital image is a color digital image comprising C components ce { 1, C} , for example C=3 RGB components (for Red, Green, Blue, in French red, green, blue), the cost function L is given by :
[0043] L-R+D=R+ 1 v” V7 / 3 / c \21 3.HW ( Xnjj - X^ij ) I
[0044] with H the number of lines of the current digital image xn, W the number of columns of the current digital image -¼ X^jj a value of a pixel of a component c of the current digital image xn and ¢7 . . a value of a pixel of a component c of the reconstructed image xn.
[0045] According to a variant, the value of a weighting coefficient ^nrij corresponding to a pixel (j) belonging to an area of interest of a current digital image xn is greater than a value of a weighting coefficient ^nx.i corresponding to a pixel (4, l) not belonging to an area of interest of said current digital image xn.
[0046] This variant makes it possible to assign more bits to compress the areas of interest of a current digital image x" and fewer bits to compress the areas of said current digital image x" which are of less interest. In other words, this variant makes it possible to concentrate the majority of the bits for the compression of the areas of interest of a current digital image x" to the detriment of the rest of said current digital image xn which is of less interest, that is to say less semantic content.
[0047] Indeed, the higher the value of a weighting coefficient ^n / ,j, the higher the flow rate (first data D) and the visual quality of a reconstructed image xn. To limit the distortion caused by compression in the areas of interest of said current digital image xn, the weighting coefficients A-njj of the pixels belonging to these areas will then have significant values compared to those of the co efficient weighting of pixels not belonging to these areas.
[0048] Conversely, the lower the value of a weighting coefficient ^nj,j, the lower the flow rate (first data D) and the visual quality of said reconstructed image xn. Compression distortions of pixels not belonging to areas of interest of said current digital image xn will be present but may be troublesome because these areas do not carry the semantic content of said current digital image xn.
[0049] According to the present invention, it is therefore advantageous to assign values to the weighting coefficients associated with the pixels of a current digital image x„ as a function of the membership of these pixels in areas of interest of said current digital image x„.
[0050] According to a variant, several areas of interest being defined in a current digital image x», values of weighting coefficients ^njj corresponding to pixels, called first pixels, belonging to said areas of interest of said current digital image xn are fixed so that a distortion calculated between the first pixels of said current digital image x» and the first pixels of said reconstructed digital image xn does not exceed a threshold value and the values of weighting coefficients corresponding to pixels, called second pixels, not belonging to said areas of interest of said current digital image xn are fixed so as to distribute compression errors over said second pixels.
[0051] According to a variant, values of weighting coefficients corresponding to pixels belonging to an area of interest of a current digital image x» are fixed so that a distortion calculated between the pixels of said area of interest of said current digital image xn and the pixels of said area of interest of said reconstructed digital image xn does not exceed a threshold value.
[0052] According to a variant, the weighting coefficient values are matched with a segmentation map representing the areas of interest in a current digital image. Said segmentation map can be obtained by a method of segmenting said current digital image.
[0053] According to one variant, the weighting coefficient values are entered by a user.
[0054] According to a variant, the values of weighting coefficients are different depending on the type of area of interest, for example if one wishes to compress certain objects more than others.
[0055] Alternatively, the compression process is taken from the article by Renjie Zou et al. entitled “The Devil Is in the Details: Window-based Attention For Image Compression” (https: / / arxiv.org / pdf / 2203.08450.pdf).
[0056] [Fig.l] schematically illustrates the lossy compression process of a digital image using coding parameters determined from a neural network defined according to a particular and non-limiting example of the present invention.
[0057] The process of lossy compression of a digital image comprises an encoder 10, an entropic predictor 11 implementing a neural network and a decoder 12.
[0058] During the learning phase, a set of current digital images xn is considered. Each current digital image x« of said set is considered in turn.
[0059] The encoder 10 comprises a divider of a current digital image xn into patches (areas of the image) and a feature extractor which extracts a feature tensor from the patches of the current digital image xn. The encoder 10 also comprises a succession of processing layers. Each processing layer applies a transformation to the feature tensor at the input of this processing layer and upsamples the transformed feature tensor. The encoder 10 produces as output a feature tensor 3'- The en-tropic predictor 11 predicts a probability distribution of the feature tensor 3' at the output of the encoder 10 for current values of the parameters of the neural network. This entropic predictor 11 generates current values of entropic coding parameters also called entropic parameters ( / / .oj which correspond to parameters of a probability density capturing statistics of the feature tensor.
[0060] Alternatively, the entropic parameters ( / / , c) correspond to a mean and a standard deviation of a Gaussian function. The details of this entropic predictor are described in the article by David Minnen et al. entitled “Channel-wise autoregressive entropy models for learned image compression” (In ICIP, 2020).
[0061] According to other variants, the entropic parameters correspond to the parameters of a mixture of Gaussian functions.
[0062] According to other variants, the entropic parameters ( / / , c) correspond to the parameters of a mixture of Gaussian, Laplacian and logistic functions.
[0063] The characteristic tensor 3' is quantized by a quantizer 13 and a quantized characteristic tensor is obtained at the output of this quantizer 13. The quantized characteristic tensor is then encoded by an encoder 14.
[0064] According to a variant, the characteristic tensor is encoded in binary by the arithmetic coding method (in English Arithmetic coding) taking as input the current values of the entropic parameters ( / / , rr).
[0065] According to a variant, the characteristic tensor is encoded by an entropic coding method such as for example the ANS method (from the English Asymmetric Numeral System in French asymmetric digital system).
[0066] The quantized characteristic tensor y encoded is decoded by a decoder 15 taking as input the current values of the entropic parameters ( (y ).
[0067] According to a variant, the quantized characteristic tensor encoded there is decoded into binary by the arithmetic coding method.
[0068] According to a variant, the quantized characteristic tensor encoded therein is decoded by an entropic decoding method such as, for example, the ANS method.
[0069] A 5' characteristic tensor is then obtained.
[0070] According to a variant, the characteristic tensor ÿ is equal to the quantized characteristic tensor y decoded at the output of the decoder 15.
[0071] According to a variant, the characteristic tensor ? is obtained by adding the quantized characteristic tensor y decoded at the output of the decoder 15 to a residue r obtained by the entropic predictor 11 from the current values of the entropic parameters ( / / , cr ) • The characteristic tensor >' is presented at the input of the decoder 12. The decoder 12 is the symmetric of the encoder 10. It comprises a succession of processing layers. Each processing layer applies a transformation to the characteristic tensor at the input of this processing layer and sub-samples the transformed characteristic tensor. The decoder 12 also comprises a reconstructor which obtains a reconstructed image î from the characteristic tensor at the output of the succession of processing layers.
[0072] A cost function L is then obtained for the current values of the parameters of the neural network.
[0073] A cost function is thus obtained for each current digital image x» of said set for said current values of the parameters of the neural network.
[0074] An average value of the obtained cost functions is calculated for the current digital images of said set. Step values are determined according to a gradient descent method used to minimize the average value of the cost functions, each step value being used to modify a current value of a parameter of the neural network. The values of the parameters of the neural network are then obtained by successive iteration, at each iteration the current values of the parameters of the neural network are modified according to step values which become smaller and smaller over the iterations. The iterations (the learning phase) cease when a stopping condition is verified as explained above.
[0075] During the production phase, a digital image to be compressed is presented as input to the encoder 10. A quantized characteristic tensor and entropic parameter values ((j) are obtained from the trained entropic predictor 11 so as to minimize the cost function L. This quantized feature tensor and the values of the entropic parameters nj can be compressed by the encoder 14 and transmitted to a decoder which would then include the decoder 15 and the decoder 12. For example, the values of the entropic parameters cr) can be transmitted as metadata. A quantized and coded feature tensor is decoded by the decoder 15 and presented as input to the decoder 12. The reconstructed image is then obtained at the output of the decoder 12.
[0076] [Fig.2] illustrates a flowchart of the different stages of a compression process with losses of a digital image using parameters determined from a neural network according to a particular and non-limiting exemplary embodiment of the present invention.
[0077] The lossy compression method includes a learning phase during which the neural network is trained.
[0078] The learning phase comprises steps applied to a set of N current digital images {1; 2V}) - At least one area of interest is defined per current digital image xn.
[0079] In the first step 31, a weighting coefficient value Àu. / is obtained for each pixel ( j ) of each current digital image xn of the set of current images, the weighting coefficient value depending on the membership of the pixel ( i, j ) of a current digital image x» to a zone of interest defined in said current digital image.
[0080] In the second step 32, initial current values of the parameters of the neural network are selected.
[0081] In a third step 33, a value of a cost function L is obtained for each current digital image xn of the set of current images.
[0082] Said cost function L is defined by a sum of a first data R and a second data D. The first R and second D data are evaluated for current values of said parameters of the neural network.
[0083] In a fourth step 34, an average value of the cost functions L obtained is calculated.
[0084] In a fifth step 35, step values are determined according to a gradient descent method used to minimize the average value of the cost functions L, each step value being used to modify a current value of a parameter of the neural network.
[0085] The third, fourth and fifth steps are iterated until a stopping condition of the gradient descent method is verified.
[0086] [Fig.3] schematically illustrates a compression device with losses of a digital image using parameters determined from a neural network according to a particular and non-limiting exemplary embodiment of the present invention.
[0087] The device 200 is for example configured for the implementation of at least one of the steps of the method described with regard to [Fig. 2]. Examples of such a device 200 include, but are not limited to, a server, a computer, on-board electronic equipment such as an on-board computer of a vehicle, an electronic calculator such as an ECU (“Electronic Control Unit”), a smartphone, a tablet, a laptop. The elements of the device 200, individually or in combination, can be integrated into a single integrated circuit, into several integrated circuits, and / or into discrete components. The device 200 can be produced in the form of electronic circuits or software (or computer) modules or even a combination of electronic circuits and software modules.According to various particular embodiments, the device 200 is communicatively coupled with other similar devices or systems, for example via a communication bus or through dedicated input / output ports.
[0088] The device 200 comprises one (or more) processor(s) 201 configured to execute instructions for carrying out the steps of the method and / or for executing the instructions of the software(s) embedded in the device 200. The processor 201 may include integrated memory, an input / output interface, and various circuits known to those skilled in the art. The device 200 further comprises at least one memory 202 corresponding for example to a volatile and / or non-volatile memory and / or comprises a memory storage device which may comprise volatile and / or non-volatile memory, such as EEPROM, ROM, PROM, RAM, DRAM, SRAM, flash, magnetic or optical disk.
[0089] The computer code of the embedded software(s) comprising the instructions to be loaded and executed by the processor is for example stored in the memory 202.
[0090] According to a particular and non-limiting embodiment, the device 200 comprises a block 203 of interface elements for communicating with external devices or a remote server or equipment on board another vehicle such as, for example, sensors, navigation and / or driving assistance systems, or a near-field communication reader or a radio receiver. The interface elements of the block 203 comprise one or more of the following interfaces:
[0091] RF radio frequency interface, for example of the Bluetooth® or Wi-Fi® type, LTE (from the English “Long-Term Evolution” or in French “Evolution à long terme”), LTE-Advanced (or in French LTE-advanced);
[0092] USB interface (from the English “Universal Serial Bus” or “Universal Serial Bus”) in French);
[0093] HDMI interface (from the English “High Definition Multimedia Interface” or “High Definition Multimedia Interface” in French);
[0094] LIN interface (from the English “Local Interconnect Network”).
[0095] According to a variant, the device 200 comprises a communication interface 204 which makes it possible to establish communication with other devices via a communication channel 205. The communication interface 204 corresponds for example to a transmitter configured to transmit and receive information and / or data via the communication channel 205 such as external events taken into consideration by the device 200 for the implementation of the method of [Fig. 2]. The communication interface 204 corresponds for example to a wired network of the CAN (Controller Area Network), CAN FD (Controller Area Network Flexible Data-Rate), FlexRay (according to the ISO 17458 standard) or Ethernet (according to the ISO / IEC 802.3 standard).
[0096] According to an additional particular embodiment, the device 200 can provide output signals to one or more external devices, such as a display screen, one or more speakers and / or other peripherals via output interfaces not shown, respectively.
[0097] The present invention also relates to a vehicle, for example an automobile or more generally a land motor vehicle, comprising a sensor configured as an object detector and the device 200 of [Fig.3].
[0098] According to a variant, at least one sensor is embedded in the vehicle and said at least one sensor is configured to detect objects located in the environment of the vehicle, i.e. objects located around the vehicle within a determined radius corresponding to the associated detection field of each sensor.
[0099] The sensor(s) correspond, for example, to one or more LIDARs (from the English “Light Detection And Ranging”).
[0100] Of course, the present invention is not limited to the embodiments and other examples or variants described above but extends to all embodiments and / or variants which would have the same scope.
Claims
Claims
1. A method for lossy compression of a digital image using coding parameters determined from a neural network, said method comprising a learning phase for determining values of the parameters of the neural network which comprises the following steps for a set of current digital images: a) obtaining (31) a weighting coefficient value for each pixel of each current digital image of said set of current digital images, said weighting coefficient value depending on whether the pixel of a current digital image belongs to a defined area of interest in said current digital image; b) selecting (32) current values of the parameters of the neural network;c) obtaining (33) a value of a cost function for each current digital image of the set of current digital images, said cost function being defined by a sum of a first data item and a second data item, said first and second data items being evaluated for said current values of said parameters of the neural network, the first data item being representative of an average number of bits per pixel necessary to encode said current digital image using the current values of said parameters of the neural network and the second data item being representative of a distortion between said current digital image and a digital image reconstructed from the current values of said parameters of the neural network, said second data item being dependent on the values of the weighting coefficients; d) calculating (34) an average value of the values of the cost functions obtained;e) determining (35) step values according to a gradient descent method used to minimize the average value of the cost functions, each step value being used to modify a current value of a parameter of the neural network; steps c, d) and e) are iterated as long as a stopping condition of the gradient descent method is not verified.;
2. Method according to claim 1, for which the second data is a mean square error that measures the distortion between a current digital image and a reconstructed digital image corresponding to the current digital image.
3. Method according to claim 1 or 2, for which the value of a weighting coefficient corresponding to a pixel belonging to an area of interest of a current digital image is greater than a value of a weighting coefficient corresponding to a pixel not belonging to an area of interest of said current digital image.
4. Method according to claim 3, for which several areas of interest are defined in a current digital image, values of weighting coefficients corresponding to pixels, called first pixels, belonging to said areas of interest of said current digital image are fixed so that a distortion calculated between the first pixels of said current digital image and the first pixels of said reconstructed digital image does not exceed a threshold value and the values of weighting coefficients corresponding to pixels, called second pixels, not belonging to said areas of interest of said current digital image are fixed so as to distribute compression errors over the second pixels.
5. A method according to claim 3, wherein weighting coefficient values corresponding to pixels belonging to an area of interest of said current digital image are set so that a distortion calculated between the pixels of said area of interest of said current digital image and the pixels of said area of interest of said reconstructed digital image does not exceed a threshold value.
6. Method according to one of the preceding claims, for which the values of weighting coefficients are matched with a segmentation map representing areas of interest in a current digital image, said segmentation map can be obtained by a method of segmenting said current digital image.
7. Device (200) for lossy compression of a digital image comprising means for implementing steps of the method according to one of claims 1 to 6.
8. Vehicle comprising at least one sensor and the device (200) according to claim 7.
9. Computer program product comprising instructions adapted for executing the steps of the method according to one of claims 1 to 6, when the computer program is executed by at least one processor.
10. A computer-readable recording medium on which is recorded a computer program comprising instructions for carrying out the steps of the method according to one of claims 1 to 6.
