MULTISPEKTRALFILTER

Description

Technical field of the invention

[0001] The invention relates to the field of multi-spectral imaging, in particular the invention relates to a method of configuring a multi-spectral filter for a multi-spectral sensor and the definition of its structure. Prior art

[0002] An image sensor consists of a multitude of photosites arranged in a grid across its entire surface. The photosites capture the intensity of the light to which they are exposed without distinguishing between wavelengths and therefore colors. To differentiate colors, the image sensor applies a multispectral filter. This filter filters a single wavelength range at each photosite, typically characterized by its average or central wavelength.

[0003] Thus, the filter detects a unique color for each photosite. This color corresponds to the spectral integration of the received light passing through the filter with the spectral transmission of the filter specifically placed on said photosite, and corresponds to the average or central wavelength of said filter. Subsequent image processing of the data acquired by the photosites allows for the reconstruction of a multi-color image of an observed scene.

[0004] Des filtres multispectraux sont connus des documents Pierre-Jean Lapray ET AL: "Multispectral Filter Arrays: Recent Advances and Practical Implementation",Sensors, vol. 14, no. 11, 17 novembre 2014 (2014-11-17), pages 21626-21659; EP 3 187 909 A1; US 2015 / 144770 A1; EP 3 557 625 A1; WO 2009 / 120928 A2; US 2018 / 170093 A1; US 2013 / 077958 A1; US 2016 / 064434 A1; US 2014 / 054448 A1 et MONNO YUSUKE ET AL: "Multispectral demosaicking with novel guide image generation and residual interpolation",2014 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP), IEEE, 27 octobre 2014 (2014-10-27), pages 645-649.

[0005] Modern multispectral filters are characterized by a basic pattern consisting of a number of photosites close to the number of wavelength ranges to be detected. For example, the Bayer filter, also known as the RGB filter, consists of a 2x2 basic pattern with two photosites detecting green, one photosite detecting red, and one photosite detecting blue. Since each photosite detects only one color, the optronic chain downstream of the sensor must interpolate the two remaining colors that were not detected for each photosite. This step is called demosaicing, demosaicing, or debayering. For multispectral applications, there are filters with a 3x3 basic pattern that allow the detection of a larger number of spectral bands, namely nine.However, this type of filter can cause problems during the debayering stage, leading to inconsistent image quality across the nine colors. This is because a sampling issue related to the optical percussive response (PSF) arises, as this response is itself linked by diffraction phenomena to the wavelength and aperture number. Thus, this type of multispectral filter does not take into account the sensor's sampling capabilities in relation to the diffraction of the optical system at the photosites, which reduces the accuracy and fidelity of the entire optical-sensor-debayering optronic chain.The present presentation aims to address these drawbacks. Summary of the invention

[0006] To this end, the present exposition presents a multispectral filter with a filtering matrix for a sensor with an array of elementary sensors or photosites, said filtering matrix comprising an elementary pattern formed from an arrangement of N elementary cells capable of filtering central wavelengths λ 1 ,..., λ k ,..., λ N , the position of the elementary cells sensitive to central wavelengths in the elementary pattern being determined such that: each wavelength being associated with an integer e 1 , ..., e k ,..., e N , chosen so that each of the products e 1 x λ 1 , ..., e k x λ k , ..., e N x λ N is essentially constant, each elementary cell sensitive to the central wavelength λ k , is positioned at okpositions in the elementary pattern such that the maximum ratio between the distance between two proximal positions of the elementary cell sensitive to said central wavelength and the associated central wavelength is minimal.

[0007] The associated central wavelength of a cell can correspond to the center of gravity weighted by the spectral transmission of the filter in question, referred to as the "center wavelength" center. In other words, the wavelengths are weighted by the spectral transmissions at the considered wavelengths.

[0008] In particular, the associated central wavelength of a cell can be determined using the following formula: ∑ 1 n λ i ∗ transmission λ k ∑ 1 n transmission λ k with transmission λk being the spectral transmission of the associated central wave λ k .

[0009] The elementary pattern may be as follows: λ 9 ± Δλ 9 λ 3 ± Δλ 3 λ 6 ± Δλ 6 λ8 ± Δλ 8 λ 3 ± Δλ 3 λ 7 ± Δλ 7 λ 9 ± Δλ 9 λ 4 ± Δλ 4 λ 7 ± Δλ 7 λ 1 ± Δλ 1 λ 5 ± Δλ 5 λ 9 ± Δλ 9 λ 6 ± Δλ 6 λ 5 ± Δλ 5 λ 8 ± Δλ 8 λ 1 ± Δλ 1 λ 5 ± Δλ 5 λ 8 ± Δλ 8 λ 2 ± Δλ 2 λ 8 ± Δλ 8 λ 6 ± Δλ 6 λ 4 ± Δλ 4 λ 4 ± Δλ 4 λ 2 ± Δλ 2 λ 6 ± Δλ 6 λ 3 ± Δλ 3 λ 6 ± Δλ 6 λ 9 ± Δλ 9 λ 7 ± Δλ 7 λ 3 ± Dλ 3 λ 8 ± Dλ 8 λ 1 ± Dλ 1 λ 7 ± Dλ 7 λ 9 ± Dλ 9 λ 7 ± Dλ 7 λ 4 ± Dλ 4 λ 5 ± Dλ 5 λ 9 ± Dλ 9 λ 6 ± Dλ 6 λ 5 ± Dλ 5 λ 8 ± Dλ 8 λ 6 ± Dλ 6 λ 5 ± Dλ 5 λ 8 ± Dλ 8 λ 6 ± Dλ 6 λ 8 ± Dλ 8 λ 2 ± Dλ 2 λ 4 ± Dλ 4 λ 9 ± Dλ 9 λ 3 ± Dλ 3 λ 1 ± Dλ 1 λ 4 ±Dλ 4 λ 9 ± Dλ 9 λ 3 ± Dλ 3 λ 1 ± Dλ 1 λ 8 ± Dλ 8 λ 6 ± Dλ 6 λ 7 ± Dλ 7 λ 9 ± Dλ 9 λ7 ± Dλ 7 λ 7 ± Dλ 7 λ 7 ± Dλ 7 λ 5 ± Dλ 5 λ 9 ± Dλ 9 λ 6 ± Dλ 6 λ 5 ± Dλ 5 λ 8 ± Dλ 8 λ 4 ± Dλ 4 λ 5 ± Dλ 5 λ 8 ± Dλ 8 λ 6 ± Dλ 6 λ 8 ± Dλ 8 λ 4 ± Dλ 4 λ 2 ± Dλ 2 λ 9 ± Dλ 9 λ 1 ± Dλ 1 λ 6 ± Dλ 6 λ 3 ± Dλ 2

[0010] With wavelengths {λ₁ - λ₂} being the following series {1100; 917; 786; 688; 611; 550; 500; 458; 423}, and Δ λ k being a predetermined wavelength difference for each wavelength.

[0011] The gap Δ λ k The wavelength can be equal to the difference between two consecutive wavelengths. λ k , λ k+1 .

[0012] The gap Δ λ k The wavelength can be equal to half the difference between two consecutive wavelengths. λ k , λ k+1 .

[0013] The gap Δ λ k The wavelength can be equal to one-quarter of the difference between two consecutive wavelengths. λ k , λ k+1 .

[0014] The present invention also relates to a method for determining a multi-spectral filtering matrix for a sensor, the method comprising: provide N central wavelengths λ 1 , ..., λk,..., λNto filter by the filtering matrix, determine an elementary pattern of the filtering matrix within the filtering matrix, the determination of the elementary motif comprising: associate an integer e 1 , ..., e k , ..., e N at each central wavelength, each integer being chosen such that the product e 1 x λ 1 , ..., e k x λ k , ..., e N x λ N that the value of each integer and the associated central wavelength is substantially constant, for each central wavelength λ k , determine ok positions of the elementary cell sensitive to said central wavelength λ k in the elementary pattern such that the ratio between the maximum distance between two closest, or proximal, positions of the elementary cell sensitive to said wavelength and the wavelength associated with said integer is minimal.

[0015] Choosing integers that ensure a nearly constant product between each integer and its associated central wavelength allows for the statistical standardization of the image sensor's transfer function. The judicious placement of ok central wavelengths λ k , This allows for consideration of the optical cutoff frequency and the spatial sampling frequency of the sensor for each wavelength. The filtering matrix determined by this process is therefore more reliable and precise because it takes into account the characteristics of the image sensor.

[0016] The filter matrix can have dimensions corresponding to a grid of photosites constituting the image sensor. The elementary pattern can be repeated in the filter matrix.

[0017] According to one embodiment, the determination of the ek positions of each of the elementary cells sensitive to the N central wavelengths λ kcan include determining a cost function Ø, said cost function being determined by the following formula: ∀ k , ∀ i k j k , ∅ = max min i k dist i k j k λ k

[0018] With k an integer between {1, ..., N}, ik , jk being integers between {1, ..., e k }, jk being different from I am, I am. ( I , jk ) being the Euclidean distance in the plane between the positions of the I th< and jk th< photosites sensitive to the central wavelength λ k And my name { distance ( ik , jk )} being for each position I the minimum distance to all positions jk said cost function incorporating pattern repetition.

[0019] This distance can be the actual distance between elementary cells.

[0020] The plurality of positions for each wavelength-sensitive elementary cell can therefore be determined by the positions for which the cost function is lowest. The products ekx λ k are thus considered constant when they satisfy the following formula: écartype e k xλ k moyenne e k xλ k ≤ A

[0021] A being chosen to be less than or equal to 15% and preferably less than 7%

[0022] The integers ok are thus chosen to satisfy this formula.

[0023] The basic pattern can be a square matrix. This reduces the computational resources required to implement the process and simplifies manufacturing.

[0024] The size of the elementary motif is determined by the following formula: ∑ 1 N e k = p 2 with p being the dimension of the elementary pattern.

[0025] The process further includes: assigning to each position of the elementary pattern a central wavelength associated with the integer arranged at said position of the elementary pattern.

[0026] The process may include, for one or more positions of the elementary pattern, a step of reassigning a first wavelength of wavelengths by a second different wavelength having a cost function at said position lower than the cost function of the first wavelength at said position.

[0027] The predetermined number N of wavelengths is greater than three and is in particular equal to nine.

[0028] This presentation also concerns a filtering matrix for a multi-spectral sensor obtained by the process as described above.

[0029] The present presentation concerns a multi-spectral sensor comprising a filter corresponding to a filtering matrix obtained by the aforementioned filtering matrix determination process.

[0030] The present presentation concerns a device for determining a filtering matrix for a multi-spectral sensor comprising a processing circuit for implementing the aforementioned process.

[0031] The present presentation concerns a computer program comprising instructions for the implementation of the aforementioned filter matrix determination process, when said instructions are executed by a processor of a processing circuit. Brief description of the figures

[0032] [ Fig. 1 ] there figure 1 represents an example of the implementation of a process for configuring a multi-spectral filter, [ Fig. 2 ] there figure 2 represents a filter obtained by the process of figure 1, in which the central wavelengths λ k are each represented by their characteristic integer ok , each central wavelength λ k being present ok times, [ Fig. 3 ] there figure 3 represents an image capture device equipped with a filter of the figure 2 . [ Fig. 4 ] there figure 4 represents the filter with the central wavelengths λ k positioned ok times, corresponding to the filter of the figure 2 . Detailed description of the invention

[0033] The 300 image capture device of the figure 4is configured to provide an image of a scene 302. The image capture device 300 comprises an optical stage 304 followed by an electronics stage 310. The optical stage 304 includes, for example, an optical lens. The electronics stage 310 includes an image sensor 308 composed of a plurality of photosites arranged in a grid over the entire surface of the sensor 308. The electronics stage 310 is configured to apply a filter 302 to the image sensor 308. The electronics stage 310 is further configured to perform the following steps: Step 312 of demosaicing, also called demosaicing or debayering, of interpolation of the data detected by the image sensor 308. Step 314 of image processing from the data interpolated during step 312. Step 316 of return of a final image representing the scene 302.

[0034] For the multispectral imaging system, sensor 300 is located at the optical focus of these systems, and the image is formed by optical stage 304, located upstream of sensor 300, from the observed scene. Filter 302 is arranged at the level of image sensor 308.

[0035] There figure 1 Figure 100 represents an example of a method for configuring a multispectral filter for an image sensor, for example, filter 302 of device 300. The image sensor can be a broadband sensor, for example, made of Si, InGaAs, InSb, or HgCdTe. The image sensor comprises a plurality of photosites arranged in a grid over its entire surface. Method 100 determines a filter matrix to be applied to the image sensor. In particular, method 100 determines a repeating elementary pattern in the filter matrix.

[0036] The process 100 includes a step 102 of receiving or determining the central wavelengths to be filtered λ 1 , ..., λ k ,..., λ N , with k being an integer and N being the number of wavelengths λ kFor example, for a silicon sensor, the wavelengths range from 380 nm to 1100 nm. In a particular embodiment, the center wavelengths can be {1100; 917; 786; 688; 611; 550; 500; 458; 423} nm. In this example, a silicon or InGaAs sensor is preferred because these materials are photosensitive to wavelengths within the range of values ​​mentioned, between 423 nm and 1100 nm. The center wavelengths can differ by + / -100 nm from the values ​​in the list {1100; 917; 786; 688; 611; 550; 500; 458; 423} nm. The spectral widths of each of the filters will be at half-height proportional to a fraction between 1 / 4 and 1 of the difference between 2 consecutive center wavelengths.Each wavelength varies within an interval defined by a wavelength deviation around said wavelength, the wavelength deviation being a fraction between 0.25 and 1 of the difference between two consecutive central wavelengths.

[0037] Thus, the minimum and maximum limits of the filters can take values ​​from those in the following table, with an uncertainty of + / -20nm: [Table 1] fraction 0.25 0.5 1 Wavelength Terminal max. Borne min Terminal max. Borne min Terminal max. Borne min 1100 1146 1054 1192 1009 1283 917 917 956 878 996 839 1074 760 786 815 757 843 729 901 672 688 710 666 732 644 776 601 611 628 594 646 577 680 542 550 564 536 578 522 606 495 500 512 489 523 477 546 454 458 468 448 477 439 497 420 423 432 414 441 406 458 388

[0038] Process 100 includes a step 104 of determining a series of integers e 1 , ..., ek , ..., e N Each integer ok is associated with a central wavelength λk. The integers ok are determined so that the product e 1 x λ 1 , ..., e k x λ k ,..., e N x λ N of each integer ok and the wavelength λ kassociated is essentially constant.

[0039] For example, integers ok are determined to satisfy the following formula: écartype e k xλ k moyenne e k xλ k ≤ A

[0040] The value A is chosen to be less than or equal to 15%. Preferably, the value A is less than 7%. The choice of integers ok allows the amount of energy folded to be standardized at each central wavelength λ k Indeed, for a signal having a bandwidth [0, fc] where fc represents the optical cutoff frequency at 1 / ( l *F), where F is the aperture number and with a sampling frequency fech, then for frequencies belonging to [0; fech], the signal is well sampled and will be faithfully reproduced, but for frequencies belonging to [fech; fc], the signal will pass through but will not be faithfully reproduced. In the latter case, it will be literally folded over the correctly reproduced portion.

[0041] This allows for the standardization, that is to say, making the ratio between the optical cutoff frequency and the spatial sampling frequency (fech k) of the sensor as equal / constant as possible for each wavelength. λ k .

[0042] Indeed, the spatial sampling frequency (fech k) of the sensor is on average equal to the inverse of the average distance between photosites sensitive to the central wavelength λ k , and therefore proportional to the number ok photosites sensitive to λ k in the elementary pattern.

[0043] Thus, the standardization of optronic chains at different wavelengths is achieved by a product ekx λ k as homogeneous as possible, or even constant.

[0044] Process 100 includes a step 106 for determining the dimension of the elementary motif. For example, the elementary motif can be square, which simplifies the implementation of process 100 in terms of the computational resources required. In this case, step 106 can be performed by solving the following formula: ∑ 1 N e k = p 2

[0045] With p being the dimension of the elementary pattern.

[0046] For the following center wavelengths: {1100; 917; 786; 688; 611; 550; 500; 458; 423} nm, the integer series can be {5; 6; 7; 8; 9; 10; 11; 12; 13} and the dimension of the elementary pattern can be 9x9.

[0047] The process 100 then includes a step 108 of determining one or more positions for each central wavelength λ k The central wavelengths are positioned ( l, m) in the elementary pattern such that the ratio between the distance to the nearest photosite sensitive to the same central wavelength and the central wavelength is minimal.

[0048] Step 108 involves minimizing a cost function Ø. For this, a cost function Ø is determined for each integer ok according to the following formula: ∀ k , ∀ i k j k , ∅ = max min i k dist i k j k λ k

[0049] With k an integer between {1, ..., N}, ik , jk being integers between {1, ..., e k }, I being different from jk , distance ( I , jk ) being the Euclidean distance in the plane between the positions of the I th< and jk th< photosites sensitive to the central wavelength λ k And my name { distance ( ik , jk )} being for each position I the minimum distance to all positions jk .

[0050] Thus, a central wavelength λ k is assigned for each position ( l, m ) of the elementary pattern.

[0051] The distance distance ( I , jk ) can be determined between two photosite positions sensitive to the central wavelength λ k within the same elementary motif or between a first photosite position λ k in a first elementary motif and a second photosite position λ k in a second elementary pattern adjacent to the first elementary pattern. Thus, the cost function Ø incorporates the repetition of the pattern.

[0052] The process 100 may include a step of generating the filter matrix by repeating the elementary pattern within the filter matrix. The filter matrix has dimensions corresponding to the number and distribution of photosites in the image sensor. A central wavelength λ kis thus assigned to each photosite of the image sensor.

[0053] Step 108 can result in a plurality of distinct elementary patterns. In this case, process 100 can include a step of selecting an elementary pattern from among the elementary patterns resulting from step 108. This selection can be random.

[0054] Process 100 may include an elementary pattern optimization step. This step includes the reassignment of a position sensitive to a central wavelength. λ k at a second central wavelength λ k in the elementary pattern. The elementary pattern after reassignment has a lower cost function than the initial elementary pattern. Reassignments are chosen by observation or by trial and error. The integers associated with the central wavelengths then evolve into e'k = ok - 1 and e'h = eh + 1.

[0055] The optimization step can be performed several times, for several central wavelengths and several positions.

[0056] The process 100 further includes a step 110 of repeating the pattern determined in the filtering matrix.

[0057] As an example, the process can result in the filter matrix 200 shown in the figure 2 for the series of integers {5; 6; 7; 8; 9; 10; 11; 12; 13} associated with the series of wavelengths {1100; 917; 786; 688; 611; 550; 500; 458; 423} positioned on the filter of the figure 4 and resulting in filter 400. The filter matrix 200 comprises a plurality of elementary patterns 202 repeated in the filter matrix and obtained in step 108. The elementary pattern is repeated in the filter matrix in step 110. The integer ok arranged at position 204 of the elementary pattern is replaced by another integer e ' k enabling a lower cost function.

[0058] The 200 filter matrix obtained by the 100 process is more accurate and reliable, especially for a large number of wavelengths, i.e., a number greater than 3. In addition, such a 200 filter matrix reduces post-interpolation errors during image processing because the filter matrix takes into account the spatial sampling frequency of the sensor and its optical cutoff frequency.

[0059] The wavelengths of the 400 filter matrix correspond to the integers of the 200 filter matrix.

[0060] The image sensor can be equipped with either the 200 filter matrix or the 400 filter matrix of the figure 4to detect 9 distinct colors corresponding to the wavelengths {1100; 917; 786; 688; 611; 550; 500; 458; 423} nm. The image sensor may include an image processing module configured to interpolate the data acquired by the image sensor to form an image.

Claims

1. A multispectral filter with a filtering array for a sensor with an array of elementary sensors, said filtering array comprising an elementary pattern formed of an arrangement of N elementary cells capable of filtering central wavelengths λ1, ..., λk, ..., λN, the position of the elementary cells sensitive to the central wavelengths in the elementary pattern being determined so that: each wavelength being associated with an integer e1, ..., ek, ..., eN, selected so that each of the products e1x λ1, ..., ekx λk,..., eNx λN is substantially constant, each elementary cell sensitive to the central wavelength λk is positioned at ek positions in the elementary pattern so that the maximum ratio between the distance between two proximal positions of said elementary cell sensitive to the central wavelength and the associated central wavelength is minimal.

2. Multispectral filter according with a filtering matrix for a sensor with a matrix of elementary sensors, said filtering matrix comprising an elementary pattern formed of an arrangement of N elementary cells adapted to filter central wavelengths λ1, ..., λk, ..., λN, the position of the elementary cells sensitive to the central wavelengths in the elementary pattern being determined to form the following elementary pattern: λ9 ± Δλ9 λ3 ± Δλ3 λ6 ± Δλ6 λ8 ± Δλ8 λ3 ± Δλ3 λ7 ± Δλ7 λ9 ± Δλ9 λ4 ± Δλ4 λ7 ± Δλ7 λ1 ± Δa,1 λ5 ± Δλ5 λ9 ± Δλ9 λ6 ± Δλ6 λ5 ± Δλ5 λ8 ± Δλ8 λ1 ± Δλ1 λ5 ± Δλ5 λ8 ± Δλ8 λ2 ± Δλ2 λ8 ± Δλ8 λ6 ± Δλ6 λ4 ± Δλ4 λ4 ± Δλ4 λ2 ± Δλ2 λ6 ± Δλ6 λ3 ± Δλ3 λ6 ± Δλ6 λ9 ± Δλ9 k7 ± Δλ7 λ3 ± Δλ3 λ8 ± Δλ8 λ1 ± Δλ1 λ ± Δλ7 λ9 ± Δλ9 λ ± Δλ7 λ4 ± Δλ4 λ4 ± Δλ4 k5 ± Δλ3 λ9 ± Δλ9 λ6 ± Δλ6 λ5 ± Δλ5 λ8 ± Δλ8 λ6 ± Δλ6 λ5 ± Δλ5 λ8 ± Δλ8 λ6 ± Δλ6 λ8 ± Δλ8 λ2 ± Δλ2 λ4 ± Δλ4 λ9 ± Δλ9 λ3 ± Δλ3 λ2 ± Δλ2 λ1 ± Δλ1 λ4 ± Δλ4 λ9 ± Δλ9 λ3 ± Δλ3 λ1 ± Δλ1 λ8 ± Δλ8 λ6 ± Δλ6 λ ± Δλ7 λ9 ± Δλ9 λ7 ± Δλ7 λ ± Δλ7 λ7 ± Δλ7 λ3 ± Δλ3 λ9 ± Δλ9 λ6 ± Δλ6 λ5 ± Ak5 λ8 ± Δλ8 λ4 ± Δλ4 λ5 ± Ak5 λ8 ± Δ8 λ6 ± Δλ6 λ8 ± Δλ8 λ4 ± Δλ4 λ2 ± Δλ2 λ9 ± Δλ9 λ1 ± Δλ1 λ6 ± Δλ6 λ3 ± λ3 λ2 ± Δλ2 The wavelengths {λ1-λ9} being the following series {1,100; 917; 786; 688; 611; 550; 500; 458; 423}, and Δλk being a predetermined wavelength deviation for each wavelength.

3. The filter according to claim 2, wherein the wavelength deviation Δλk is equal to the difference between two consecutive wavelengths λk, λk+1.

4. The filter according to claim 2, wherein the wavelength deviation Δλk is equal to half the difference between two consecutive wavelengths λk, λk+1.

5. The filter according to claim 2, wherein the wavelength deviation Δλk is equal to a quarter of the difference between two consecutive wavelengths λk, λk+1.

6. A method (100) for determining a multispectral filtering array for a sensor, the method comprising: - providing (102) N central wavelengths λ1, ..., λk, ..., λN to be filtered by the filtering array, - determining an elementary pattern (202) of the filtering array (200) in the filtering array, the determination of the elementary pattern comprising: - associating (104) an integer e1, ..., ek,..., eN with each wavelength, each integer being selected so that each product e1x λ1, ..., ekx λk,..., eNx λN of each integer and of the associated central wavelength is substantially constant, - for each central wavelength λk, determining (108) ek positions of the elementary cell sensitive to said central wavelength in the elementary pattern so that the maximum ratio between the distance between two proximal positions of the elementary cell sensitive to said central wavelength and the associated central wavelength is minimal.

7. The method (100) according to claim 6, wherein the determination (108) of the plurality of ek positions of each elementary cell sensitive to the central wavelength λk comprises determining a cost function Ø for all of the N central wavelengths, said cost function being determined by the following formula: ∀ k , ∀ i k j k , ∅ = max min i k dist i k j k λ k With k being an integer comprised between {1, ...,N}, ik, jk being integers belonging to {1, ..., ek}, jk being different from ik, dist(ik, jk) being the Euclidean distance in the plane between the positions of the ikth and ikth photosites sensitive to the central wavelength λk and minik{dist(ik, jx)} being for each position ik the minimum distance to all of the positions jk, said cost function integrating the repetition of the pattern. the plurality of positions of each elementary cell sensitive to the central wavelength being determined by the positions (l, m) of the elementary pattern for which the cost function is the lowest.

8. The method (100) according to claim 6 or 7, wherein the products ekx λk meet the following formula: standard deviation e k xλ k average e k xλ k ≤ A where A is selected less than or equal to 15% and preferably less than 7%.

9. The method (100) according to one of claims 6 to 8, wherein the elementary pattern is a square array.

10. The method (100) according to one of claims 6 to 8, wherein the dimension of the elementary pattern is determined by the following formula: ∑ 1 N e k = p 2 with p being the dimension of the elementary pattern.

11. The method (100) according to one of claims 6 to 10, wherein the predetermined number N of wavelengths is greater than three.

12. A filtering array (200) for a multispectral sensor obtained by the method (100) according to one of claims 6 to 11.

13. A multispectral sensor comprising equipped with a filter corresponding to a filtering array obtained by the method (100) according to one of claims 6 to 11.

14. A device for determining a filtering array for a multispectral sensor including a processing circuit for implementing the method (100) according to one of claims 6 to 11.

15. A computer program including instructions for the implementation of the method (100) according to one of claims 6 to 11, when said instructions are executed by a processor of a processing circuit.