FIELD OF THE INVENTION

The present invention relates to an image sensor comprising a Bayer matrix.

STATE OF THE ART

It is known from the state of the art an image sensor comprising a Bayer matrix and a photodetector arranged downstream of the Bayer matrix.

In known manner, a Bayer matrix comprises a plurality of optical filters arranged in a checkerboard pattern and sensitive to different wavelengths. Thus, a Bayer matrix conventionally comprises at least the following nine optical filters:

• a reference optical filter, configured to eliminate or attenuate in an optical signal received a first band of wavelengths comprising a first wavelength λ f but to allow a second band of wavelengths comprising a second length to pass. wave λ 2 ,

• eight optical filters adjacent to the optical reference filter, four of which are configured to pass the first band of wavelengths into the received optical signal, and to eliminate or attenuate the second band of wavelengths.

The photodetector arranged downstream of the Bayer matrix makes it possible to reconstruct an image, each pixel of the image corresponding to one of the optical filters of the Bayer matrix. Thus, a reference pixel of the image is specifically associated with the aforementioned reference filter.

The reconstructed image carries information in the two spectral bands to which the Bayer matrix is ​​sensitive.

In particular, the reference pixel of the image is decked out with a color produced from a first item of information relating to the first band of wavelengths, and from a second item of information relating to the second band of lengths d. 'wave.

The second information is supplied directly by the photodetector, since the optical reference filter allows this second band to pass.

However, the first information cannot be obtained in the same way, since the reference optical filter eliminates or at least attenuates this first band. Also, this first information is obtained indirectly by combining the information relating to the first band of lengths in pixels adjacent to the reference pixel, these adjacent pixels being associated with adjacent optical filters allowing the first band of wavelengths to pass.

In the literature, the step consisting in reconstructing, for each pixel from the information available in said pixel or in its vicinity, the data of each of the wavelength bands, is conventionally called “demosaicing” (or

“Debayerisation” by neologism). When the wavelength bands are close, for example in a visible application in which a Bayer matrix having red-green-blue (VB) filters is used, known demosaicing algorithms implement an analysis of a local gradient and neglect the dispersion of the optical impulse response caused by the Bayer matrix (also called “spreading function”, and in English “Optical Point Spread Function”, abbreviated as OPSF).

There are two main reasons for explaining the reasons for this approach. The theory shows that the optimal optical impulse response (ie limited by diffraction) is an Airy spot whose first dark ring has a diameter equal to 2.44 * lambda * N, where lambda is considered wavelength and N is number opening of the system (N = focal length / diameter of the entrance pupil).

In the case of a Bayer matrix in the visible domain, the average wavelengths of the RBV spectral bands are relatively close, and, moreover, the

OPSF (linked to the optical qualities of the optical filters forming the Bayer matrix) are far from the diffraction limit. This leads to OPSFs associated with the RGB spectral bands sufficiently similar to neglect their differences.

However, in other applications, the OPSFs at the different wavelength bands to which the Bayer matrix is ​​sensitive may be significantly different; in a first example, when the length bands are moved away from each other; in a second example, when the optical qualities of the optical filters constituting the Bayer matrix are such that the similarity of the OPSFs is no longer a realistic approximation.

The demosaicing implemented for Bayer matrices operating in the visible domain, and neglecting the OPSF differences, can therefore no longer be used to reconstruct an image.

These difficulties are encountered most particularly when we have λ 1 / λ 2 = 1/2, where λ 1 is included in the first band and λ 2 is included in the second band. Such a relationship can occur in particular when the first band is chosen in the infrared band “SWIR” (“Short-wavelength infrared”) and the second band chosen in the infrared band “MWIR” (“Mid-wavelength infrared”).

In such a configuration, similar prior art optical qualities lead to similar OPSFs (for example close to diffraction) with respect to the wavelength band for each band of wavelength. wave considered, and therefore at a ratio of 2 between the radial dimensions of the OPSFs. From a spatial energy distribution point of view, this means that for the same point energy at infinity, the illuminated surface at the level of the photodetector array is approximately four times smaller in the first band than in the second band.

A first way to get around the problem would consist in adjusting the number of apertures N to the first wavelength band by increasing it, for example by a factor of 2 compared to the N 1n itiai which would have been chosen for the configuration. optimal for the second wavelength band. However, this would have the consequence of spreading the radial dimension of the OPSFs by a factor of 2 without making them similar (in the direction of diffraction) from one wavelength band to another, of reducing by a factor 4 the energy density deposited per unit area at the photodetector and increase the exposure time by a factor of 4 compared to the configuration with the
A second way would be to design an optical system having an aperture number 1 ^ for the first wavelength band and an aperture number N 2 for the second wavelength band, such that 1 ^ = ^ / 2. The OPSFs of the two length bands would thus be similar (in the sense of diffraction) but the required exposure time of the first wavelength would increase by a factor of 4 compared to the 1 ^ = ^ configuration.

A third way would consist in deliberately introducing optical aberrations into the design of the optical system, in a preferred manner for the rays of the first spectral band, in order to make the OPSFs of the two wavelength bands similar. The drawback of this solution is that it leads to OPSFs which are not spatially co-localized in the imaging field.

DISCLOSURE OF THE INVENTION

An object of the invention is to allow the reconstruction of a multispectral image by an image sensor using a Bayer matrix sensitive to two bands of wavelengths and generating a significant optical impulse response dispersion.

Another object of the present invention is to provide an image sensor making similar the reconstructed images in two such wavelength bands so that the image processing unit can evaluate the intensities received in each spectral band by each. element of the image plane.

Yet another object of the present invention is to modulate the compromise between the number of apertures, the exposure time and the size of the photodetector used in an image sensor using a Bayer matrix.

There is thus proposed, according to a first aspect of the invention, an image sensor comprising:

• an optical system for receiving an optical signal, the optical system comprising a pupil,

• a Bayer matrix located at the image focal plane of the optical system, the Bayer matrix comprising: an optical reference filter configured to eliminate or attenuate in the optical signal received a first band of wavelengths and to allow the signal to pass optical received a second band length of wavelengths, and moreover eight optical filters adjacent to the optical reference filter,

• a phase mask arranged on the pupil and configured to project at least 98% of the energy of the optical signal transported in the first band of wavelengths and 98% of the energy of the optical signal transported in the second band of wavelengths selectively on the reference optical filter and on at least one adjacent optical filter which is configured to pass in the received optical signal the first band of wavelengths.

The image sensor according to the first aspect of the invention can also include the following optional characteristics, taken alone or in combination when this is technically possible.

In a first embodiment, the phase mask is configured to project at least 98% of said energy selectively on the reference optical filter and two of the adjacent optical filters; each of the two adjacent optical filters is further configured to pass in the received optical signal the first band of wavelengths and to eliminate or attenuate in the received optical signal the second band of wavelengths.

At least 49% of said energy can then be projected onto the reference optical filter and at most 24.5% of said energy preferably be projected onto each of the two adjacent optical filters.

Furthermore, the phase mask can have a flat surface and include a boss in the form of a triangular prism projecting from the flat surface, the boss preferably comprising two free faces connected by a rounded edge.

In a second embodiment, the phase mask is configured to project at least 98% of said energy selectively on the reference optical filter and four of the adjacent optical filters; each of the four adjacent optical filters is then configured to allow the first band of wavelengths to pass through the received optical signal.

The phase mask according to this second embodiment can be configured to project at least 32% of said energy onto the reference optical filter.

In addition, the phase mask may have a flat surface and moreover a boss projecting from the flat surface, the boss having four opposite flat faces in pairs and preferably connected by rounded corners.

Moreover, this boss may have an invariant shape by rotation of 90 degrees about an axis normal to the flat surface.

In a third embodiment, the phase mask is configured to project at least 98% of said energy selectively onto the reference optical filter B1 and the eight adjacent optical filters.

The phase mask can then be configured to project at least 29% of said energy onto the reference optical filter.

Further, the phase mask can be configured to project at least 58% of said energy onto four of the adjacent optical filters, and each of the four adjacent optical filters can be configured to eliminate or attenuate in the received optical signal a second band of lengths d. wave different from the first band.

The phase mask may have a planar surface and includes a central boss projecting from the planar surface over a first height, as well as an annular boss also projecting from the planar surface to a second height less than the first height, the boss annulus extending around and away from the central boss.

The phase mask may have a planar surface with a boss protruding from the flat surface, wherein the boss is tapered or has a tapered concave shape.

Each boss can be of revolution about an axis normal to the flat surface.

According to a second aspect of the invention, there is proposed an image acquisition method by an image sensor comprising an optical system for receiving an optical signal, a Bayer matrix located at the image focal plane of the optical system, the Bayer matrix comprising a reference optical filter configured to eliminate or attenuate in a received optical signal a first band of wavelengths, and eight optical filters adjacent to the reference optical filter, the method being characterized by a projection, by a mask phase arranged on a pupil of the optical system, of at least 98% of

the optical signal energy transported in the first wavelength band and 98% of the optical signal energy transported in the second wavelength band selectively over the reference optical filter and over at least one of the eight adjacent optical filters, the Bayer matrix being located at the image focal plane of the phase mask.

DESCRIPTION OF FIGURES

Other characteristics, aims and advantages of the invention will emerge from the following description, which is purely illustrative and non-limiting, and which should be read with reference to the appended drawings in which:

“FIG. 1 is a schematic side view of certain components of an image sensor according to one embodiment of the invention.

Figures 2 and 3 are partial front views of a Bayer matrix.

FIGS. 4 and 5 are respectively a side view and a front view of a phase mask according to a first embodiment of the invention.

“FIG. 6 represents two phase shift curves of an optical signal passing through the phase mask according to the first embodiment of the invention.

FIG. 7 represents the point spreading function of the phase mask according to the first embodiment of the invention.

Figure 8 shows two point spread curves associated with the phase shift curves shown in Figure 6.

FIG. 9 is a partial front view of a Bayer matrix, also showing an energy distribution projected onto this Bayer matrix by the phase mask according to the first embodiment of the invention.

Figures 10 and 11 are respectively a perspective view and a front view of a phase mask according to a second embodiment of the invention.

FIG. 12 is a partial front view of a Bayer matrix, also showing an energy distribution projected onto this Bayer matrix by the phase mask according to the second embodiment of the invention.

FIGS. 13 and 14 are respectively a side view and a front view of a phase mask according to a third embodiment of the invention.

FIG. 15 is a side view of a phase mask according to a fourth embodiment of the invention.

Figure 16 is a perspective view of a phase mask according to a fifth embodiment of the invention.

FIG. 17 is a partial front view of a Bayer matrix, also showing an energy distribution projected onto this Bayer matrix by the phase mask according to the fifth embodiment of the invention.

In all of the figures, similar elements bear identical references.

DETAILED DESCRIPTION OF THE INVENTION

With reference to FIG. 1, an image sensor 1 comprises an optical system 2a represented here by its principal planes H and H ', a pupil of the optical system 2b, a phase mask 2c, a Bayer matrix 4, a photodetector 6 and an image processing unit 8.

The Bayer 4 matrix is ​​conventional. It comprises a plurality of optical filters arranged on a rectangular grid of dimensions N x M along a plane defined by two axes: an X axis and a Y axis perpendicular to the X axis (only the Y axis being visible in FIG. 1 ).

The photodetector 6 is arranged downstream of the Bayer matrix 4.

The conventional photodetector 6 typically comprises a photosensitive surface covered by the Bayer matrix 4.

The photodetector 6 is adapted to generate, for each optical filter of the Bayer matrix 4, an electrical signal from an optical signal which has passed through the optical filter.

The image processing unit 8, also conventional, is configured to apply to the various electrical signals generated by the photodetector 6 an image processing known in itself making it possible to construct an image consisting of N x M pixels, each pixel corresponding to one of the optical filters of the Bayer matrix 4.

The image processing unit 8 comprises for example at least one processor executing an image processing program configured for this purpose.

The optical system 2a and the phase mask 2c have an optical axis Z.

The optical system 2a is arranged upstream of the Bayer matrix 4, so that the Bayer matrix 4 is at the image focal plane of the optical system 2a. The image focal plane of the optical system 2a is therefore the (X, Y) plane.

The phase mask 2c is positioned upstream of the Bayer matrix 4 and preferably arranged on a pupil of the optical system 2b.

The pupil 2b may equally well be the entrance pupil, that is to say the ' image of the diaphragm d ' opening through the part of the optical system 2a located upstream, or the exit pupil, that is i.e. the ' image of the diaphragm d ' opening through the part of the optical system located downstream, or intermediate pupil, that is to say the ' image of the aperture stop for only part of the elements of optical system. Those skilled in the art will recognize the usual definitions and will know that the entrance, exit and intermediate pupils are the conjugates of each other by the elements of the optical system which separate them.

The phase mask 2c is arranged so as to be able to transmit the optical signal of the different wavelength bands. In the following, a first band comprising the wavelength λ 1 and a second band comprising the wavelength λ 2 will be considered .

The phase mask 2c is designed so as to introduce into the pupil a set of 0 DM path differences , homogeneous at optical thicknesses, usually given in metric “meter” units.

The phase mask 2c is designed so that the set of 0 DM path differences seen by each wavelength band correspond to two phase shifts Δ λ1 and Δ λ2 , usually measured in radians. In the particular case where λ 1 / λ 2 = 1/2, it is obvious to a person skilled in the art that Δ λ2 = Δ λ1 / 2.

It should be noted that each of the phase shifts can be implemented modulo (2 * n), and therefore be written Δ λη = Δ λη Γ6ί + ^ 2 * π, with k integer. By default in the embodiments below, k is taken equal to 0 and .delta λη = .delta λη Γ6ί ·

With reference to FIG. 2, the Bayer matrix 4 comprises at least the following nine optical filters:

• a reference optical filter B1, configured to eliminate or attenuate a first band of wavelengths in a received optical signal, but to allow a second band of wavelengths to pass,

• eight optical filters B2, B3 adjacent to the optical reference filter B1.

The nine filters B1, B2, B3 are aligned so as to form a 3 x 3 checkerboard, for example square when the optical filters are themselves square.

The eight adjacent optical filters B2, B3 include four adjacent first optical filters B2 configured to pass the first band of wavelengths in the received optical signal, and to eliminate or attenuate the second band of wavelengths. The first four adjacent filters B2 are connected to the filter B1: they are arranged respectively to the left, to the right, above and below the filter B1.

The eight adjacent optical filters B2, B3 furthermore comprise four second adjacent optical filters B3 configured to allow the second band of wavelengths to pass in the optical signal received, and to eliminate or attenuate the first band of wavelengths. They are therefore of the same type as the reference filter B1. The fourth

second adjacent filters B2 are arranged at the four corners of the 3 x 3 checkerboard formed by the nine filters B1, B2, B3.

For example, the first band is chosen from the infrared band “SWIR” (“Short-wavelength infrared”) and the second band is chosen from the infrared band “MWIR” (“Mid-wavelength infrared”).

Phase mask 2c is configured to project at least 98% of the energy carried by an optical signal in the first band of wavelengths and 98% of the energy carried by an optical signal in the second band of length d wave selectively on the optical reference filter B1 and on at least one of the eight adjacent optical filters B2 and B3. By convention, 98% of this total energy is referred to below as “useful energy”. The area of ​​the Bayer matrix 4 in which this useful energy is projected is inscribed in the circle drawn in dotted lines in Figure 3.

The image of a point by an optical system not being a point but a diffraction figure which, at best, is an Airy spot whose dimensions are imposed by the wavelength of the radiation, we usually consider its size defined at the first minimum of this profile (starting from the center of the spot, also called 1st dark ring). The integral of the energy on this spot-centered disc corresponds to approximately 98% of the total energy.

There is shown in Figures 4 and 5 a first embodiment of phase mask 2c to implement such a projection.

The phase mask 2c according to this first embodiment is in the form of a phase plate 10 comprising an upstream surface 12 and a downstream surface 14 opposite to the upstream surface 12. The two surfaces 12, 14 are oriented so that the optical signal enters the plate 10 via the upstream surface 12, and leaves it via the downstream surface 14. The downstream surface 14 is arranged opposite the Bayer matrix 4.

The two surfaces 12, 14 are plane, and extend in respective planes parallel to the plane (X, Y) of the Bayer matrix 4.

The phase mask 2c comprises a boss 16 projecting from the upstream surface 12 planar. The boss 16 has four opposite flat faces two by two 18a-18d. The faces 18a-18d are interconnected by roundings 20. These roundings have the effect of avoiding uncontrolled diffraction peaks.

Ultimately, the boss has the overall shape of a pyramid 18a-18d with four faces, the edges and apex of which would be machined to form the roundings 20.

The boss 16 has an invariant shape by rotation of 90 degrees around the Z axis normal to the upstream flat surface 12.

The boss 16 is not of revolution: it has a first width measured parallel to the X or Y axis, and a second width measured along a diagonal axis of the X and Y axes and which is greater than the first width.

The dimensions of the boss are adapted so that the phase mask 2c shifts an optical signal in the first band of wavelengths (comprising λ) according to the following phase shift d, a function of the incidence coordinates (x, y) of this optical signal on the upstream surface of phase mask 2c:

d = A l (x, y) / n = max (l - 0.80 * sin (x 2 + y 2 ) + 0.84 * cos (x + y 4 ), 0)

As shown in Figure 6, the phase shift curve in the plane (X, Z) or in the plane (Y, Z) is different from the phase shift curve induced by the phase mask 2c in a plane defined by the Z axis but rotated 45 ° around the Z axis with respect to the plane (X, Z).

The point spreading function (impulse response, in English PSF for “Point Spread Function”) of the phase mask 2c for the optical signal in the first wavelength band is represented in FIGS. 7 and 8. It is noted that this function defines a cross in a plane parallel to the plane (X, Y).

As a result, an optical signal having passed through the phase mask 2c can be selectively projected onto a formed cross region selectively covering the reference optical filter B1 and the first four adjacent optical filters B2, as shown in Fig. 9. In this As an embodiment, the other four adjacent optical filters B3 therefore do not receive the useful energy (therefore no energy transported by the optical signal in the first band of wavelengths.

Preferably, the phase mask 2c (in particular its boss 16) has a shape adapted so that at least 50% of the useful energy is projected onto the reference optical filter B1, which corresponds to at least 49% of l total energy carried by the optical signal in the first band of wavelengths.

Preferably, the phase mask 2c (in particular its boss 16) has a shape suitable for each of the first four optical filters B2 to receive at least 16.7% of the useful energy (therefore corresponding to at least 16.3% of total energy).

The phase mask is for example made of ZnS, with an optical index n ¾1 . M ~ = 2.265 for λ 1 = 2.1 μτη and λ 2 = 4.2 μτη.

Let R be the distance between the point considered of the phase mask 2c and the center of the pupil, that is to say the distance between the point considered of the phase mask 2c and the optical axis Z illustrated in FIG. 1 , and E the thickness of the phase plate between its two surfaces 12 and 14, measured parallel to the Z axis.

By way of example, when the pupil has a radius equal to 10 mm, the thickness of the phase mask 2c is chosen as follows:

• For R e [0mm; 1 .626mm]: Ε + 2.493μηη

where Φ ΒΜ = 2.493 * (2.265-1) = 3.15μηη

o Δ ¾1 / π = 3 = 1 and Δ ¾2 /π=1.5

• For R e [1 .626mm; 6.335mm]: E

o Φο Μ = 0μηη

o Δ λ1 / π = 0 and Δ λ2 / π = 0

• For R e [6.335mm; 10mm]: Ε + 0.831 μηη

where Φ ΒΜ = 0.831 * (2.265-1) = 1 .04μηη

o Δ λ1 / π = 1 and Δ λ2 /π=0.5

The operation of the image sensor 1 is as follows.

An optical signal is received by image sensor 1. This optical signal enters the phase mask 2c via its upstream surface 12 and in particular its boss 16. The optical signal passes through the plate 10 and leaves it via the downstream surface 14. During this passage, the optical signal undergoes a phase shift. The optical signal is then projected onto the Bayer matrix 4, selectively onto the reference optical filter B1 and the first four adjacent optical filters B2.

The reference optical filter B1 attenuates or eliminates, in the optical signal that it receives, the first band of wavelengths comprising λ 1 f but allows the second band of wavelengths comprising λ 2 to pass . The optical signal thus filtered is detected by the photodetector 6, converted into an electrical signal, then transmitted to the image processing unit 8.

Moreover, each of the first optical filters B2 attenuates or eliminates, in the optical signal that it receives, the second band of wavelengths comprising λ 2 , but allows the first band of wavelength comprising λ 1 to pass . The optical signal thus filtered is detected by the photodetector 6, converted into an electrical signal, then transmitted to the image processing unit 8.

The five electrical signals generated by the photodetector 6 are combined according to a method known per se by the image processing unit to produce a color associated with a pixel of an image.

In the above, it has been assumed that the Bayer matrix only includes 9 optical filters arranged in a 3 x 3 checkerboard pattern. Of course, the Bayer matrix can include many more optical filters. The image sensor then comprises not a single mask 2, but as many masks 2 as there are optical filters comprising 8 neighbors in the Bayer matrix 4, and consequently capable of being considered as reference optical filters B1.

The preceding steps are thus implemented for each optical filter capable of being considered as a reference optical filter, so as to obtain at the output of the image processing unit 8, a plurality of colors of pixels, the pixels forming a complete picture.

There is shown in Figures 10 and 11 a second embodiment of phase mask 22 to implement a projection of 98% of the energy of an optical signal transported in the first band of wavelengths selectively over the reference optical filter B1 and on at least one adjacent optical filter. This phase mask 22 can replace the phase mask 2c in the sensor shown in Figure 1.

The phase mask 22 differs from the phase mask 2c according to the first embodiment in that it comprises two bosses 24, 26 each projecting from the upstream flat surface 12, and not just one: a central boss 24 and a peripheral boss 26.

The central boss 24 has a cylindrical shape of revolution about the Z axis. The central boss 24 has a plane top and parallel to the upstream surface 12. The central boss has a height measured parallel to the Z axis between the upstream surface. and its vertex equal to a value H1.

The peripheral boss 26 has an annular or crown shape, of revolution around the Z axis. This peripheral boss 26 is located around and at a distance from the central boss 24. The peripheral boss 26 has a plane top and parallel to the upstream surface. 22. The peripheral boss 26 has a height measured parallel to the Z axis between the upstream surface and its apex equal to a value H2 less than H1.

The phase mask 22 according to this second embodiment is suitable for projecting the useful energy throughout the circle, the perimeter of which is drawn in dotted lines in FIG. 12. In other words, unlike the phase mask 2c according to the first embodiment , energy is received by the nine optical filters B1, B2, B3 when the phase mask 22 is used.

Preferably, the shape of the bosses 24, 26 is adjusted so that the phase mask 22 projects at least one third of the useful energy on the reference optical filter B1 (therefore at least 29% of the total energy transported by the optical signal in the first band of wavelengths).

Preferably, the shape of the bosses 24, 26 is adjusted so that the phase mask 22 projects at least 60% of the useful energy on the four adjacent optical filters (which represents about 58% of the total energy). For example, the Z axis passes through the center of the reference filter, so that each of the four adjacent optical filters receives 15% of the useful energy. The four other adjacent optical filters each receive 2.5% of the useful energy in this case.

The phase mask 22 according to the second embodiment has the advantage of being simpler to manufacture than the mask 2 according to the first embodiment.

There is shown in Figures 1 3 and 14 a third embodiment of phase mask 28 to implement a projection of 98% of the energy of an optical signal transported in the first band of wavelengths selectively over the reference optical filter B1 and the eight adjacent optical filters B2, B3, just like the mask 22 according to the second embodiment.

Like the mask 2 according to the first embodiment, this phase mask 28 comprises a single boss 30 projecting from the upstream planar surface 12. This boss 30 however has a conical shape of revolution about the Z axis. this boss is triangular in a plane perpendicular to the upstream surface, as shown in figure 12.

FIG. 15 shows a fourth embodiment of a phase mask 2c making it possible to implement a projection of 98% of the energy of an optical signal transported in the first band of wavelengths selectively on the optical filter. reference B1 and the eight adjacent optical filters B2, B3, just like the masks 22 and 28 according to the second and the third embodiment.

Like the mask according to the first embodiment, this phase mask 2c comprises a single boss 34 each projecting from the upstream flat surface. This boss 34 however has a tapered concave shape. The profile of this boss in a plane perpendicular to the upstream surface is not triangular but curved, as shown in figure 14.

FIG. 16 shows a fifth embodiment of a phase mask 36 making it possible to implement the expected projection of the useful energy.

The phase mask 36 has a profile identical to that of the mask 2c shown in FIG. 5. The mask 36 indeed has a boss 38 with a generally triangular profile projecting from the upstream surface 12. However, the boss 38 has a general shape of triangular prism and not pyramidal.

This triangular prism is defined by a generatrix parallel to one of the two X or Y axes of the Bayer matrix.

The boss comprises two free faces 40, 42 rectangular connected by an edge located at a distance from the upstream surface 12 (these two free faces corresponding to two of the three faces of the prism, the last face being in the plane of the upstream surface 12) .

Preferably, the two free faces 40, 42 are not connected by an edge but by a rounding, and / or each of the two free faces 40, 42 is connected to the flat surface 12 by a rounding. These roundings have the effect of avoiding uncontrolled diffraction peaks.

When the phase mask 36 is used in the image sensor 1 shown in FIG. 1, this mask 36 projects an incident optical signal on only the reference optical filter B1 and two adjacent first optical filters B2 which are preferably connected to the filter. B1. These three filters are aligned parallel to one of the two X or Y axes.

Preferably, the phase mask 36 is suitable for at least 50% of the useful energy on the reference optical filter B1 (therefore at least 49% of the total energy transported by the optical signal in the first band of lengths d 'wave). The two optical filters B2 immediately to the left and to the right (or respectively at the top and at the bottom) of the filter B1 each receive at least 25% of the useful energy, as CLAIMS

1. Image sensor including:

• an optical system (2a) for receiving an optical signal, the optical system comprising a pupil (2b),

• a Bayer matrix (4) located at the image focal plane of the optical system (2a), the Bayer matrix (1) comprising: a reference optical filter (B1) configured to eliminate or attenuate a first band in the optical signal received wavelengths and to allow a second band length of wavelengths to pass in the received optical signal, and moreover eight optical filters (B2, B3) adjacent to the reference optical filter (B1),

the image sensor being characterized in that it further comprises

• a phase mask (2c, 22, 28) arranged on the pupil (2b) and configured to project at least 98% of the energy of the optical signal transported in the first band of wavelengths and 98% of the energy of the optical signal transported in the second band of wavelengths selectively over the reference optical filter (B1) and over at least one adjacent optical filter (B2) which is configured to pass in the received optical signal the first band of wavelengths.

2. The image sensor of claim 1, wherein the phase mask is configured to project at least 98% of said energy selectively onto the reference optical filter (B1) and two of the adjacent optical filters, and wherein each of the two adjacent optical filters is configured to pass in the optical signal received the first band of wavelengths and to eliminate or attenuate in the optical signal received the second band of wavelengths.

3. Image sensor according to claim 2, wherein at least 49% of said energy is projected onto the optical reference filter (B1) and wherein at most 24.5% of said energy is preferably projected onto each of the. two adjacent optical filters.

4. Image sensor according to one of claims 1 to 3, wherein the phase mask (2c) has a flat surface and comprises a boss (38) in the form of a triangular prism projecting from the flat surface, the boss preferably comprising two free faces connected by a rounding.

5. The image sensor of claim 1, wherein the phase mask (2c) is configured to project at least 98% of said energy selectively onto the reference optical filter (B1) and four (B2) of the adjacent optical filters. , and wherein each of the four adjacent optical filters (B2) is configured to pass in the received optical signal the first band of wavelengths.

6. The image sensor of claim 5, wherein the phase mask (2c) is configured to project at least 32% of said energy onto the optical reference filter (B1).

7. Image sensor according to one of claims 5 to 6, wherein the phase mask has a flat surface and further comprises a boss projecting from the flat surface, wherein the boss has four opposite flat faces two to two and preferably connected by rounding.

8. The image sensor of claim 7, wherein the boss has a shape invariant by rotation of 90 degrees about an axis normal to the planar surface.

9. The image sensor of claim 1, wherein the phase mask is configured to project at least 98% of said energy selectively onto the optical reference filter.

(B1) and the eight adjacent optical filters.

10. The image sensor of claim 9, wherein the phase mask is configured to project at least 29% of said energy onto the optical reference filter (B1).

11. The image sensor of one of claims 9 to 10, wherein the phase mask is configured to project at least 58% of said energy onto four of the adjacent optical filters, and wherein each of the four adjacent optical filters is configured to eliminate or attenuate in the received optical signal a second band of wavelengths different from the first band.

12. Image sensor according to one of claims 9-1 1, wherein the phase mask has a flat surface and comprises a central boss projecting from the flat surface over a first height, and an annular boss forming also protruding

from the planar surface to a second height less than the first height, the annular boss extending around and away from the central boss.

13. Image sensor according to one of claims 9 to 11, wherein the phase mask has a flat surface and comprises a boss projecting from the flat surface, wherein the boss is conical or has a tapered concave shape.

14. Image sensor according to one of claims 12 to 13, wherein each boss is of revolution about an axis normal to the flat surface.

15. Image acquisition method by an image sensor comprising an optical system (2a) for receiving an optical signal, a Bayer matrix (4) located at the image focal plane of the optical system, the Bayer matrix (4). ) comprising an optical reference filter (B1) configured to eliminate or attenuate in the received optical signal a first band of wavelengths, and eight optical filters (B2, B3) adjacent to the optical reference filter (B1), the method being characterized by a projection, by a phase mask (2c, 22, 28) arranged on a pupil (2b) of the optical system (2a), of at least 98% of the energy of the optical signal transported in the first band of wavelengths and 98% of the energy of the optical signal carried in the second band of wavelengthswave selectively on the reference optical filter (B1) and on at least one of the eight adjacent optical filters (B2, B3), the Bayer matrix (4) being located at the image focal plane of the phase mask (2).illustrated in FIG. 17.