VIEW SYNTHESIS

Abstract

A method for synthesizing an image of a view from images of N (N>2), implemented by an image synthesis device. The method includes: projecting, to a position corresponding to the image of the view to be synthesized, N depth maps associated with the N views, respectively; for at least one given pixel of at least one projected depth map, for which a depth value has been associated on completion of projection, modifying the depth value of the at least one given pixel if an item of reliability information associated with the depth value is at a certain value, the modification using the depth value of a pixel whose position corresponds to that of the at least one given pixel, in at least one other projected depth map, which generates at least one modified projected depth map.

Description

1. FIELD OF THE INVENTION

The present invention relates in general to the field of image synthesis. The present invention applies more particularly to the synthesis of uncaptured intermediate viewpoints, on the basis of images of a plurality of 2D (two-dimensional) views, over 360°, 180°, etc., which are captured in order to generate an immersive video, such as in particular a 360°, 180°, etc., video.

The invention may in particular, but not exclusively, be applied to the video decoding implemented in current AVC (abbreviation for Advanced Video Coding) video decoders and their MVC (abbreviation for Multiview Video Coding) extensions, 3D-AVC, MV-HEVC, 3D-HEVC, etc.

2. PRIOR ART

In a context of immersive video, i.e. in which the viewer has the sensation of being immersed in a 3D (three-dimensional) scene, the scene is conventionally captured by a set of cameras. These cameras may be:

• • of 2D type, in order to capture a particular angle of the scene, and/or
  • of 360°, 180° or other type, in order to capture all of the scene over 360 degrees, 180 degrees or other about the camera.

The images of such captured views are conventionally coded, and then decoded by the terminal of the viewer. However, in order to deliver an adequate quality of experience, and therefore visual quality and a good degree of immersion, displaying the captured views is not enough. The images of a multitude of views, referred to as intermediate views, have to be calculated, on the basis of the decoded images of the views.

The images of these intermediate views are calculated by means of a view “synthesis” algorithm. A synthesis algorithm is capable, on the basis of the images of N views, with N≥2, of synthesizing an image of an intermediate viewpoint located anywhere in space. The image of a given view from among the N comprises a texture component and a depth map that indicates the distance separating the various elements of the scene from the camera that has captured the image of this view. Thus, the image of an intermediate view obtained by means of synthesis likewise comprises a texture component synthesized on the basis of the N texture components of the images of the N views and a depth map synthesized on the basis of the N depth maps of the images of the N views.

For a given view from among the N, the depth map is either captured or calculated on the basis of the N texture components. However, in both cases, such a depth map may contain numerous errors which are related to the capture, related to the calculation, or related to the compression of the images of the N views.

Such errors therefore inevitably have an impact on the image of a view synthesized on the basis of the image of each of these N views, which notably decreases the performance of current synthesis algorithms.

One well-known synthesis algorithm of the type mentioned above is for example the VSRS (View Synthesis Reference Software) algorithm. This algorithm implements a projection of the N depth maps to a position corresponding to the image of the view to be synthesized in the video to be generated, in order to determine a single synthesized depth map. This depth map is then used to recover the texture information associated with each pixel of the N texture components.

One drawback of this projection lies in the fact that some pixels are missing in the N projected depth maps, since they do not correspond to any of the pixels of the N depth maps before projection. This lack of correspondence may result either from a lack of precision in the pixels contained in the N depth maps used for the projection or because the missing pixels correspond to regions that are not visible in the images of the N views (occlusion) containing the N depth maps.

Such projection errors therefore inevitably have an impact on the image of a view that is synthesized on the basis of the image of each of these N views.

The errors generated in the initial calculation of a depth map are also dependent on the quantization of this information, i.e. on the number of values that this depth map may take, as well as on the level of compression applied, manifesting in the depth map as two types of defects:

• • blurring if the errors are small,
  • gross defects if these errors are substantial.

As the difference between what the camera that has captured the image of a view can see and what the virtual camera can see increases, synthesis defects become increasingly substantial, and transition from blurring to gross defects, the latter absolutely having to be avoided so as not to negatively affect the sensation of immersion of the viewer.

The techniques of the prior art do not take into account the fact that the available depth maps, intended to be projected in the synthesis of the image of an intermediate view, feature errors. As a result, the depth maps obtained on completion of the projection of the available depth maps also contain errors.

3. AIM AND SUMMARY OF THE INVENTION

One of the aims of the invention is to rectify drawbacks of the aforementioned prior art.

To that end, one subject of the present invention relates to a method for synthesizing an image of a view on the basis of images of N (N≥2) views, implemented by an image synthesis device, characterized in that it comprises the following:

• • projecting, to a position corresponding to the image of the view to be synthesized, N depth maps associated with the N views, respectively, such a method being characterized in that it comprises the following:
  • for at least one given pixel of at least one projected depth map, for which a depth value has been associated on completion of projection, modifying the depth value of said at least one given pixel if an item of reliability information associated with the depth value is at a certain value, the modification using the depth value of a pixel whose position corresponds to that of said at least one given pixel, in at least one other projected depth map, which generates at least one modified projected depth map.

Taking into account the fact that a conditional modification is applied to one or more projected pixels of at least one of the projected depth maps, the invention advantageously makes it possible to correct errors in said at least one projected depth map. For a given pixel of such a projected depth map, these projection errors may result:

• • in quantization noises introduced during the digital quantization, over a plurality of bits, of the depth value of each pixel of the depth map of the image of a view, on the basis of which the projected depth map has been obtained,
  • in errors introduced during the compression of the depth map by means of a coder of HEVC, 3D-HEVC, etc. type,
  • in errors introduced during the projection of the depth map of the image of the view into real space.

All of the pixels of a projected depth map are not mandatorily modified. They are modified if an item of reliability information, assigned to each of these pixels, is at a certain value.

Additionally, such a modification uses the depth value of a pixel whose position corresponds to that of a given pixel, in another projected depth map, such a depth value being considered relevant for correcting the erroneous depth value of the given pixel.

Thus, by virtue of the invention, a given projected depth map is marred by many fewer errors than a projected depth map of the prior art. This results in a very clear improvement in the quality of the image of a view synthesized on the basis of a plurality of images of views, when at least one of these images is associated with a depth map that contains errors before and/or after projection.

According to one embodiment of the invention, the modification uses a weighting of the depth value of the pixel whose position corresponds to said at least one given pixel, in said at least one other projected depth map.

Such an embodiment make it possible to assign a higher or lower importance to a pixel whose position corresponds to that of the given pixel, in another projected depth map, which will, as a result, have a larger or smaller effect on the modification of the depth value of the given pixel.

According to one embodiment of the invention, the confidence level of a pixel of a given depth map is calculated as a variation in the depth value of said pixel, said variation corresponding to a projection error in terms of allowed number of pixels.

Such a calculation of the confidence level advantageously makes it possible to take into account the real projection quality of said at least one other depth map, in addition to the positioning distance between the camera that has captured the image of the view, for which the projection of the depth map has generated the given projected depth map, and the camera that has captured the image of said at least one other view, for which the projection of the depth map has generated the other projected depth map.

According to one embodiment of the invention, the confidence level of a pixel is weighted by a coding parameter of the image of the view with which the depth map that contains the pixel is associated.

Such an embodiment makes it possible to refine the calculation of the confidence level of a given pixel by taking into account the compression quality level, such as, for example, the value of the quantization step that has been used during the coding of the image of the view with which the depth map that contain said pixel is associated, or else the position of this image in the coding hierarchy.

According to one embodiment of the invention, the modification of the depth value of the given pixel consists in replacing said depth value with a value that is calculated on the basis of said depth value and of the depth value of the pixel whose position corresponds to that of said at least one given pixel in said at least one other projected depth map, said depth values each being weighted by their respective confidence level.

According to one embodiment of the invention, the item of reliability information is generated as follows:

• • determining, in the set of N projected depth maps, for one and the same position as that of the given pixel, which pixel has the maximum depth value and which pixel has the minimum depth value,
  • calculating the difference between the maximum and minimum depth values,
  • comparing the calculated difference with a threshold,
  • generating an item of reliability information, the value of which is dependent on the result of the comparison.

According to one embodiment of the invention, the item of reliability information is advantageously calculated in such a way that the modification that is applied to the value of the given pixel of the given depth map and that is conditional on this item of reliability information results only in blur-type artefacts in the image of the synthesized view.

According to one embodiment of the invention, the item of reliability information is generated as follows:

• • determining, in the set of N projected depth maps, for one and the same position as that of the given pixel, which pixel has the maximum depth value and which pixel has the minimum depth value,
  • calculating a difference between the depth value of said given pixel and the determined minimum depth value,
  • comparing the calculated difference with a threshold,
  • generating an item of reliability information with respect to the determined minimum depth value, the value of which is dependent on the result of the comparison,
  • calculating another difference between the determined maximum depth value and the depth value of said given pixel,
  • comparing the other calculated difference with said threshold,
  • generating an item of reliability information with respect to the determined maximum depth value, the value of which is dependent on the result of the comparison.

According to one embodiment, the item of reliability information for a given pixel of a given projected depth map is advantageously quantized over two levels:

• • a first level that takes into account the difference between the depth value of the given pixel in the projected depth map and the minimum depth value determined for the N projected depth maps, for the N pixels whose position corresponds to that of the given pixel,
  • a second level that takes into account the difference between the maximum depth value determined for the N projected depth maps, for the N pixels whose position corresponds to that of the given pixel, and the depth value of the given pixel in the projected depth map.

In this way, for a given pixel of a projected depth map, according to the two items of reliability information associated with the given pixel, the respective depth values of the pixels whose position corresponds to that of the given pixel in the N projected depth maps may be selected in two different ways to modify the depth value of the given pixel. The modification of the depth value of the given pixel is thus made more precise.

According to one embodiment of the invention, the comparison threshold is equal to the mean of the N variations in the depth value of each pixel whose position corresponds to that of said given pixel in their respective depth map.

Such an embodiment for calculating the comparison threshold make it possible to optimize the decrease in blur-type artefacts in the image of the synthesized view.

According to one embodiment of the invention, the comparison threshold is equal to the mean of the N variances in the depth value of each pixel whose position corresponds to that of said given pixel in their respective depth map.

Such an embodiment for calculating the comparison threshold makes it possible to optimize the decrease in blur-type artefacts in the image of the synthesized view, while taking into account the intrinsic quality of the N projected depth maps.

The various abovementioned embodiments or implementation features may be added, independently or in combination with one another, to the synthesis method defined above.

The invention also relates to a device for synthesizing an image of a view on the basis of images of N (N≥2) views, such a synthesis device being characterized in that it comprises a processor that is configured to implement the following:

• • projecting, to a position corresponding to the image of the view to be synthesized, N depth maps associated with the N views, respectively,
  • for at least one given pixel of at least one projected depth map, for which a depth value has been associated on completion of projection, modifying the depth value of said at least one given pixel if an item of reliability information associated with said depth value is at a certain value, said modification using the depth value of a pixel whose position corresponds to that of said at least one given pixel, in at least one other projected depth map, which generates at least one modified projected depth map.

Such a synthesis device is notably capable of implementing the aforementioned synthesis method, according to any one of its aforementioned embodiments.

The invention also relates to a method for decoding a data signal representative of a set of images of N (N≥2) coded views, comprising the following:

• • decoding the images of the N coded views, producing a set of images of N decoded views,
  • synthesizing an image of a view on the basis of the set of images of N decoded views in accordance with the aforementioned synthesis method, according to any one of the aforementioned embodiments.

The invention also relates to a device for decoding a data signal representative of a set of images of N (N≥2) coded views, such a decoding device comprising a processor that is configured to implement the following:

• • decoding the images of the N coded views, producing a set of images of N decoded views,
  • synthesizing an image of a view on the basis of said set of images of N decoded views in accordance with the aforementioned synthesis method, according to any one of the aforementioned embodiments.

The invention also relates to a computer program comprising instructions for implementing the synthesis method or the decoding method integrating the synthesis method according to the invention, according to any one of the particular embodiments described above, when said program is executed by a processor.

This program can use any programming language, and be in the form of source code, object code, or of code intermediate between source code and object code, such as in a partially compiled form, or in any other desirable form.

The invention also targets a recording medium or information medium that is readable by a computer, and comprises instructions of a computer program, such as mentioned above.

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, for example a CD-ROM or a microelectronic circuit ROM, or else a magnetic recording means, for example a USB key or a hard disk.

Moreover, the recording medium may be a transmissible medium such as an electrical or optical signal, which may be conveyed via an electrical or optical cable, by radio or by other means. The program according to the invention may in particular be downloaded from an Internet-type network.

Alternatively, the recording medium may be an integrated circuit in which the program is incorporated, the circuit being adapted to execute or to be used in the execution of the aforementioned synthesis or decoding method.

4. BRIEF DESCRIPTION OF THE DRAWINGS

Other features and advantages will become more clearly apparent from reading several preferred embodiments, given purely by way of illustrative and non-limiting examples, and described below with reference to the appended drawings, in which:

FIG. 1 shows the main actions performed by the synthesis method according to one embodiment of the invention,

FIG. 2 shows an exemplary image used in the synthesis method of FIG. 1,

FIGS. 3A to 3C show an exemplary calculation of a confidence level used in the synthesis method of FIG. 1,

FIG. 4 shows an exemplary calculation of an item of reliability information used in the synthesis method of FIG. 1,

FIG. 5 shows a synthesis device implementing the synthesis method of FIG. 1,

FIGS. 6A and 6B show exemplary arrangements of the synthesis device of FIG. 5, in the case where the images being used for the image synthesis have been decoded beforehand.

5. DESCRIPTION OF THE GENERAL PRINCIPLE OF THE INVENTION

The invention primarily proposes a scheme for synthesizing an image of an intermediate view on the basis of a plurality of images of, respectively, a plurality of views, each view representing, at the current time, a 3D scene according to a given position or a given viewing angle.

For each image of a view in the plurality of images, the depth map of the image of the view is projected conventionally to a position corresponding to the image of a view to be synthesized.

The invention is characterized by the application of a conditional modification of the depth value of each given pixel in a given projected depth map. Such a modification makes it possible to compensate for the errors in the depth values which may be due to:

• • errors introduced during the calculation of the depth values of the depth map of the image of a view, on the basis of which the given projected depth map has been obtained,
  • errors introduced during the compression of data of the image of the view,
  • errors introduced during the projection of the depth map of the image of the view into real space.

6. EXEMPLARY SYNTHESIS SCHEME IMPLEMENTATIONS

A method for synthesizing an image of a view on the basis of images of a plurality of views is described hereinafter, such a method being able to be used in or with any type of current video decoders AVC and HEVC and their extensions (MVC, 3D-AVC, MV-HEVC, 3D-HEVC, etc.), inter alia.

With reference to FIG. 1, such a synthesis method uses N images I₁, I₂, . . . , I_j, . . . , I_N of N views, respectively, with 1≤j≤N and N≥2, the plurality of views representing a 3D scene according to, respectively, a plurality of viewing angles or a plurality of positions/orientations. Conventionally:

• • the image I₁ comprises a texture component T₁ and a depth map D₁,
  • the image I₂ comprises a texture component T₂ and a depth map D₂,
  • . . . ,
  • the image I_j comprises a texture component T_j and a depth map D_j,
  • . . . ,
  • the image I_N comprises a texture component T_N and a depth map D_N.

For a given image I_j, as shown in FIG. 2:

• • its texture component T_j comprises Q (Q≥1) points p1_j, p2_j, . . . , pQ_j each assigned a corresponding texture value t1_j, t2_j, . . . , tQ_j,
  • its depth map D_j comprises the Q points p1_j, p2_j, . . . , pQ_j each assigned a corresponding depth value d1_j, d2_j, . . . , dQ_j.

In S1 in FIG. 1, the depth maps D₁, D₂, . . . , D_j, . . . , D_N are projected to a position corresponding to an image I_sth of a view to be synthesized.

Such a projection is implemented by a projection algorithm, for example of DIBR (abbreviation for Depth Image Based Rendering) type.

On completion of such a projection, N projected depth maps D₁^v, D₂^b, . . . , D_j^v, . . . , D_N^v are obtained. A given depth map D_j from among the N comprises the Q points p1_j, p2_j, . . . , pQ_j each assigned a corresponding depth value d1^v_j, d2^v_j, . . . , dQ^v_j.

Such depth values are not systematically correct given in particular:

• • quantization noise introduced during the digital quantization over a plurality of bits of each of the depth values d1_j, d2_j, . . . , dQ_j, and/or
  • in the case where the depth map D_j has undergone compression by a coder of HEVC, 3D-HEVC, etc. type, errors introduced during this compression, and/or
  • errors introduced during the projection of the depth map D_j into real space.

In a manner known per se, the N projected depth maps D₁^v, D₂^v, . . . , D_j^v, . . . , D_N^v are associated, respectively, with N attribute maps A₁, A₂, . . . , A_j, . . . , A_N.

For any given projected depth map D_j^v (1≤j≤N), an attribute map indicates for each point/pixel p1_j to pQ_j of the depth map D_j, whether their projection is valid or not.

In the case for example of the point p1_j of the depth map D_j:

• • an attribute value A1_j=V1 is associated with the point p1_j of the projected depth map D_j^v to indicate that no depth value has been assigned to the point p1_j,
  • an attribute value A1_j=V2 is associated with the point p1_j of the projected depth map D_iv to indicate that a depth value has been assigned to the point p1_j.

The absence of a depth value attributed to a point/pixel of a projected depth map may occur during the projection S1 of this depth map, on completion of which no depth value could be assigned to the point resulting from the projection. This point without an associated depth value then constitutes a “hole” region.

In this example, each attribute value of an attribute map A₁, A₂, . . . , . . . , A_N is binary. To that end, V1=0 and V2=1 or vice versa. According to another example, the values of an attribute map A₁, A₂, . . . , A_j, . . . , A_N are integers within the range [0, +∞].

In S2, in FIG. 1, for at least one given projected depth map D_j^v, a conditional modification is applied to the depth values d1^v_j, d2^v_j, . . . , dQ^v_j.

Such a conditional modification takes place as follows.

In S21, for a given pixel pi_j (1≤i≤Q) of the projected depth map D_j^v, it is determined whether the attribute Ai_j associated with the pixel pi_j is at a first value V1 or at a second value V2.

If the attribute Ai_j has a value of V1, the conditional modification method is not implemented and the following pixel of the projected depth map D_j^v is then considered.

If the attribute Ai_j has a value of V2, in S22, it is determined whether an item of reliability information Fi_j, associated with the pixel pi_j, is at a first value V3 or at a second value V4.

According to one non-exhaustive exemplary embodiment, V3=0 and V4=1 or vice versa.

If the item of reliability information Fi_j is at the first value V3, the depth value di^v_j of the pixel pi_j of the projected depth map D_j^v is not modified.

If the item of reliability information Fi_j is at the second value V4, the depth value di^v_j of the pixel pi_j of the projected depth map D_j^v is modified in S23. According to the invention, the depth value di^v_j is modified according to a depth value of a pixel located in the same position as that of the pixel pi_j, in at least one other projected depth map D₁^v, D₂^v, . . . , D_N^v. Thus, the depth value is modified according to:

• • the depth value di^v₁ associated with the pixel pi₁ of the projected depth map D₁^v, and/or,
  • the depth value di^v₂ associated with the pixel pi₂ of the projected depth map D₂^v, if the depth value di^v₂ exists (Ai₂=V2) and/or,
  • . . . ,
  • the depth value di^v_N associated with the pixel pi_N of the projected depth map D_N^v if the depth value di^v_N exists (Ai_N=V2).

According to the invention, the modification S23 uses a weighting of the depth values d1^v_j, d2^v_j, . . . , dQ^v_j. Thus:

• • the depth value di^v₁ is weighted by a coefficient Ci₁,
  • the depth value di^v₂ is weighted by a coefficient Ci₂,
  • . . . ,
  • the depth value di^v_N is weighted by a coefficient Ci_N.

As will be described further below in the description:

• • the coefficient Ci₁ corresponds to a confidence level that has been attributed to the depth value di^v₁ of the pixel pi₁ of the projected depth map D^v₁,
  • the coefficient Ci₂ corresponds to a confidence level that has been attributed to the depth value div2 of the pixel pi2 of the projected depth map D^v₂,
  • . . . ,
  • the coefficient Ci_N correspond to a confidence level that has been attributed to the depth value di^v_N of the pixel piN of the projected depth map D^v_N.

Of course, a confidence level is attributed only if the given pixel in a given projected depth map has a depth value that corresponds to that of the pixel whose position is the same in the corresponding depth map yet to be projected.

On completion of the modification S23, a modified depth value d1_j^v of the pixel pi_j of the projected depth map D_j^v is obtained according to the following relationship:

• • dl_j^v=fonction ((Ci₁, di^v₁), (Ci₂, di^v₂), . . . , (Ci_N, di^v_N)), on condition that Ai₁=V2, Ai₂=V2, . . . , Ai_N=V2.

The conditional modification S2 described above is applied to the depth value of each of the pixels p1_j, p2_j, . . . , pQ_j of the projected depth map Dj^v and is iterated for all of the projected depth maps D₁^v to DN^v.

On completion of the conditional modification S2, the following are obtained:

• • at least one modified projected depth map D_j^v,
  • an unmodified projected depth map D₁^v or a modified projected depth map ₁^v,
  • an unmodified projected depth map D₂^v or a modified projected depth map D₂^v,
  • . . . ,
  • an unmodified projected depth map D_N^v or a modified projected depth map D_N^v.

In a manner known per se, in S3, the depth maps D₁^v or D₁^v, D₂^v or D₂^v, . . . , D_j^v, . . . , D_N^v or D_N^v are interpolated with the texture components T₁, T₂, . . . , T_N, respectively, for example using a texture backward warping algorithm, delivering an image I_sth of a synthesized view.

The image synthesis method applies equally to uncoded, coded or decoded images I₁ to I_N. In the case where the images I₁ to I_N have been decoded, the synthesis method may form part of the decoding method and be implemented by the decoder, or else be implemented in a synthesis device placed at the output of the decoder.

One embodiment of the calculation of the confidence level mentioned above will now be described, with reference to FIGS. 3A and 3B.

All of the depth maps D₁ to D_N are considered.

In S10, a depth map is selected, for example the depth map D_j.

In S11, a pixel pi_j of the depth map D_j is selected.

In S12, the pixel pi_j is projected on the projected depth map D_j^v, in a position identical to that of the pixel pi_j in the depth map D_j. During the projection S12, a variation of the depth value di_j of the pixel pi_j is carried out until reaching a maximum depth value di_jmax, such that:

Δdi_j=di_jmax−di_j<E_proj

where E_proj is an allowed projection error of the pixel pi_j on the projected depth map D_j^v.

According to one exemplary embodiment, E_proj is the projection deviation of the pixel pi_j in number of pixels, for example 3 pixels.

In S13, the confidence level Ci_j is calculated according to the relationship:

Ci _j =g(Δdi_j)

where g represents a predefined mathematical function, such as for example a power function or an exponential function.

The application of a function to Δdi_j thus makes it possible to give greater importance to the depth map located closest to the view to be synthesized.

Operations S11 to S13 are applied to all of the pixels of the selected depth map D_i so as to attribute a respective confidence level C1_j, C2_j, . . . , CQ_j to each of these pixels.

On completion of these operations, with reference to FIG. 3C, a map Ci_j of confidence levels is attributed in relation to the projected depth map D_j^v.

All of operations S10 to S13 are then applied to each of the depth maps D₁ to D_N.

According to one embodiment, in the case where the images I₁ to I_N have been coded, the calculation of a confidence level Ci_j is also dependent on at least one coding parameter parcomp associated with the image I_j. Such a coding parameter is for example the value of the quantization step that was used during the coding of the image I_j, or the position of this image in the coding hierarchy.

More generally, the calculation of the confidence level Ci_j may take into account all of the parameters of the image I_j (intrinsic and extrinsic parameters), apart from the texture, namely:

• • the real quality of the projection and not only the positioning distance between two cameras that each capture a given image from among the N images I₁ to I_N,
  • the digital quantization level: a depth map quantized on 8 bits, in which only a quarter of the dynamic range is therefore used, has a confidence level that is lower than the confidence level of this depth map quantized on 32 bits, since the variation in depth Δdi_j giving a displacement by 3 pixels will be smaller,
  • the one or more compression parameters of the image I_i (quantization step, temporal structure),
  • the intrinsic quality the depth map D_j, i.e. a small deviation between the depth value at each point/pixel of the depth map D_j and the real depth value at each corresponding point in the 3D scene,
  • the importance given to the depth map D_j, linked to the position of the image I_j with respect to the image I_sth to be synthesized.

One embodiment of a calculation of the item of reliability information mentioned above will now be described with reference to FIG. 4.

All of the projected depth maps D₁^v à DN^b are considered.

In S100, the pixels p1₁, p1₂, . . . , p1_N that are located in the same position (the first pixel of the first row from the left), in each of the projected depth maps D₁^v, D₂^v, . . . , D_j^v, . . . , D_N^v are selected.

In S101, it is determined which pixel, from among the pixels p1₁, p1₂, . . . , p1_N, has the maximum depth value d1^v_max.

In S102, it is determined which pixel, from among the pixels p1₁, p1₂, . . . , p1_N, has the minimum depth value d1^v_min.

During operations S101 and S102, if no depth value has been found in relation to a given pixel in a projected depth map, this projected depth map is not considered. As already explained above, the absence of depth value may occur during the projection S1 (FIG. 1) of each depth map D₁ à D_N, on completion of which all of the pixels of the synthesized image lath are not necessarily associated with a depth value. These pixels without an associated depth value constitute “hole” regions.

In S103, the difference Δd1^v=d1^v_max-d1^v_min is compared with a predefined threshold depth_TH.

If Δd1^v=d1^v_max-d1^v_min≥depth_TH, in S104, an item of reliability information F1 is set to a first value V3, such that for example V3=0, to indicate that the depth value of the pixels p1₁, p1₂, . . . , p1_N which have been projected to the same position, respectively in the projected depth maps D₁^v, D₂^v, . . . , D_j^v, . . . , D_N^b, is not reliable.

As a variant, the comparison S104 is Δd1^v=d1^v_max-d1^v_min>depth_TH.

If Δd1^v=d1^v_max-d1^v_min<depth_TH, in S105, an item of reliability information F1 is set to a second value V4, such that for example V4=1, to indicate that the depth value of the pixels p1₁, p1₂, . . . , p1_N which have been projected to the same position, respectively in the projected depth maps D₁^v, D₂^v, . . . , D_j^v, . . . , D_N^v, is reliable.

As a variant, the comparison S105 is Δd1^v=d1^v_max-d1^v_min≤depth_TH.

The threshold depth_TH is a threshold for differentiating between the foreground and the background of the 3D scene. The threshold depthm is dependent for example on the variations in depth value Δd1₁, Δd1₂, . . . , Δd1_j, . . . , Δd1_N that were calculated in S12 (FIG. 3A) to generate the confidence level C1₁ of the pixel p1₁, the confidence level C1₂ of the pixel p1₂, . . . , the confidence level C1_j of the pixel p1_j, . . . , the confidence level C1_N of the pixel p1_N.

According to one particular example,

$dept{h}_{TH}=\frac{{\sum }_{j=1}^N\Delta d{1}_j}{N}$

According to another particular example,

$dept{h}_{TH}=\sqrt{\frac{{\sum }_{y=0}^{\mathrm{height}}{\sum }_{x=0}^{\mathrm{width}}{\sum }_{j=1}^N{\left(D_j^v\left(x,y\right)-\stackrel{\_}{D^v}\left(x,y\right)\right)}^2}{\left(\mathrm{width}\right)*\left(\mathrm{height}\right)}}$

with

$\stackrel{\_}{D^V}\left(x,y\right)=\frac{{\sum }_{j=1}^N{D}_j^v\left(x,y\right)}{N}$

where (x, y) are the coordinates of the pixels/points of each depth map.

Operations S100 to S103 and S104 or S105 are iterated for the pixels p2₁, p2₂, . . . , p2_N that are located in the same position (for example the second pixel of the first row from the left), in each of the projected depth maps D₁^v, D₂^v, . . . , D_j^v, . . . , D_N^v, and so on up to the pixels pQ₁, pQ₂, . . . , pQ_N that are located in the same position (for example the last pixel of the last row from the left), in each of the projected depth maps D₁^v, D₂^v, . . . , D_j^v, . . . , D_N^v.

7. Particular Embodiments of the Synthesis Method

Three embodiments of the synthesis method described above will now be described.

According to these three embodiments, an image 1st of a synthesized view is obtained on the basis of images I₁, I₂, I₃, I₄ of four views, respectively.

With reference to FIG. 1, the depth maps D₁, D₂, D₃, D₄ of each image I₁, I₂, I₃, I₄, respectively, are projected, in S1, to a position corresponding to an image I_sth of a view to be synthesized. Four projected depth maps D₁^v, D₂^v, D₃^v, D₄^v are then obtained.

The following are provided as input to the synthesis method:

• • the four projected depth maps D₁^v, D₂^v, D₃^v, D₄^v obtained in S1,
  • the corresponding four attribute maps A₁, A₂, A₃, A₄ obtained in S1,
  • the predefined threshold depth_TH.

7.1 First Embodiment

In order to implement the conditional modification operation S2 of FIG. 1, the method performs the calculation, at each pixel of each of the projected depth maps D₁^v, D₂^v, D₄^v, of its respective confidence level, as illustrated in FIG. 3A.

All of the depth maps D₁ to D₄ are considered.

In S10, a depth map is selected, for example the depth map D₁.

In S11, a pixel pi, of the depth map D₁ is selected.

In S12, the pixel pi₁ is projected on the projected depth map D₁^v, in a position identical to that of the pixel pi₁ of the depth map D₁. During the projection S12, a variation of the depth value di₁ of the pixel pi₁ is carried out until reaching a maximum depth value di_1max, such that:

Δdi₁=di_1max−di_1≤E_proj

where E_proj is an allowed projection error of the pixel pi₁ on the projected depth map D₁^v.

According to one preferred exemplary embodiment, E_proj is the projection deviation of the pixel pi₁ in number of pixels, for example 3 pixels. It is considered that an error of 3 pixels at most primarily results in blur-type artefacts.

The calculation of Δdi₁ is iterative. To that end, the depth value di₁ is increased up to the depth value di_1max, for which a projection deviation E_proj of 3 pixels is obtained.

In S13, the confidence level Ci₁ is calculated according to the relationship Ci₁=(Δdi₁)² to give greater importance to the image of the view closest to the image I_sth of the view to be synthesized.

Operations S11 to S13 are applied to all of the pixels of the selected depth map D₁ so as to attribute a respective confidence level C1₁, C2₁, . . . , CQ₁ to each of these pixels.

All of operations S10 to S13 are then applied to each of the depth maps D₂, D₃, D₄.

The calculation of such a confidence level takes into account all of the parameters of an image of a view (texture aside), i.e. the real quality of the projection of its depth map, the quantization level of this image, and not only the positioning distance between two cameras that have captured two images from among the images I₁ to I₄.

The calculation of such a confidence level may also take as input data from a video coder, like for example a quantization step.

In order to implement the conditional modification operation S2 of FIG. 1, the method also performs the calculation of the item of reliability information, at each pixel of each of the projected depth maps D₁^v, D₂^v, D₃^v, D₄^v, as illustrated in FIG. 4.

All of the projected depth maps D₁^v to D₄^v are considered.

In S100, the pixels p1₁, p1₂, . . . , p1_N that are located in the same position (the first pixel of the first row from the left), in each of the projected depth maps D₁^v, D₂^v, D₃^v, D₄^v are selected.

In S101, it is determined which pixel, from among the four pixels p1₁, p1₂, p1₃, p1₄, has the maximum depth value d1^v_max.

According to the first embodiment:

• • d1^v_max=Max ((A1₁, D₁^v), (A1₂, D₂^v), (A1₃, D₃^v), (A1₄, D₄^v))

In S102, it is determined which pixel, from among the four pixels p1₁, p1₂, p1₃, p1₄, has the minimum depth value d1^v_min.

According to the first embodiment:

• • d1^v_min=Min ((A1₁, D₁^v), (Al₂, D₂^b), (A1₃, D₃^v), (A1₄, D₄^v))

In operations S101 and S102, A1₁ is the attribute relating to the projection of the pixel p1₁ in the projected depth map D₁^v, A1₂ is the attribute relating to the projection of the pixel p1₂ in the projected depth map D₂^v, A1₃ is the attribute relating to the projection of the pixel p1₃ in the projected depth map D₃^v, A1₄ is the attribute relating to the projection of the pixel p1₄ in the projected depth map D₄^v, and by considering that:

• • A1₁=V1 or A1₁=V2, and/or
  • A1₂=V1 or A1₂=V2, and/or
  • A1₃=V1 or A1₃=V2 and/or
  • A1₄=V1 or A1₄=V2,knowing that if A1₁=V1 (or A1₂=V1, A1₃=V1, A1₄=V1), no depth value of the pixel p1₁ (or p1₂, p1₃, p1₄) has been projected on the projected depth map D₁^v (or D₂^v, D₃^v, D₄^v) and that therefore this value does not occur in the calculation of the maximum depth value d1^v_max or of the minimum depth value d1^v_min.

In S103, the difference Δd1^v=d1^v_max-d1^v_min is compared with a predefined threshold depth_TH, according to the following relationship:

Δd1^v=d1^v_max−d1^v_min<depth_TH

with

$\mathrm{dept}{h}_{TH}=\frac{{\sum }_{j=1}^4\Delta d{1}_j}{4}$

If Δd1^v=d1^v_max-d1^v_min≥depth_TH, in S104, an item of reliability information F₁ is set to a first value V3, such that for example V3=0, to indicate that the depth value of the pixels p1₁, p1₂, p1₃, p1₄ which have been projected to the same position, respectively in the 4 projected depth maps D₁^v, D₂^v, D₃^v, D₄^v, is not reliable.

If Δd1^v=d1^v_max-d1^v_min<depth_TH, in S105, an item of reliability information F₁ is set to a second value V4, such that for example V4=1, to indicate that the depth value of the pixels p1₁, p1₂, p1₃, p1₄ which have been projected to the same position, respectively in the 4 projected depth maps D₁^v, D₂^v, D₃^v, D₄^v, is reliable.

Operations S100 to S103 and S104 or S105 are iterated for the pixels p2₁, p2₂, . . . , p2_N that are located in the same position (for example the second pixel of the first row from the left), in each of the projected depth maps D₁^v, D₂^v, D₃^v, D₄^v, and so on up to the pixels pQ₁, pQ₂, . . . , pQ_N that are located in the same position (for example the last pixel of the last row from the left), in each of the projected depth maps D₁^v, D₂^v, D₃^v, D₄^v.

With reference to the FIG. 1, for at least one given projected depth map D_j^v, a conditional modification S2 is applied to the depth values d1^v_j, d2^v_j, . . . , dQ^v_j.

Such a conditional modification S2 takes place as follows.

In S21, for a given pixel pi₁ (1≤i≤Q) of the projected depth map D₁^v, it is determined whether the attribute Ail associated with the pixel ph is at a first value V1 or at a second value V2.

If the attribute Ai₁ has a value of V1, the conditional modification method is not implemented and the following pixel of the projected depth map D₁^v is then considered.

If the attribute Ai₁ has a value of V2, in S22, it is determined whether an item of reliability information Fi₁, associated with the pixel ph, is at a first value V3 or at a second value V4.

If Fi₁=V3, the depth value di^v₁ of the pixel pi₁ of the projected depth map D₁^v is not modified.

If Fi₁=V4, the depth value di^v₁ of the pixel pi₁ of the projected depth map D₁^v is modified in S23.

According to the first embodiment, on completion of the modification S23, a modified depth value dl₁^v of the pixel pi₁ of the projected depth map D₁^v is obtained according to the following relationship:

$\stackrel{\_}{d_{l_1}^v}=\frac{{\sum }_{j=1}^4\left(C{i}_j*{\mathrm{di}}_j^v\right)}{{\sum }_{j=1}^4\left(C{i}_j\right)}$

The conditional modification S2 described above is applied to the depth value of each of the pixels p1_j, p2_j, . . . , pQ_j of the projected depth map D₁^v and is iterated for each pixel of each of the other projected depth maps D₂^v, D₃^v, D₄^v.

7.2 Second Embodiment

According to this second embodiment, the calculation of the confidence level takes place in the same way as in the first embodiment.

The calculation of the reliability information takes place in the same way as in the first embodiment up to and including S102.

The following calculation is then performed:

The pixels p1₁, p1₂, . . . , p1_N that are located in the same position (the first pixel of the first row from the left), in each of the projected depth maps D₁^v, D₂^v, D₃^v, D₄^v.

The calculation of the mean variance Var_moy of the corresponding projected depth values d1^v₁, d1^v₂, d1^v₃, d1^v₄ is performed.

If Var_moy>depth_TH, in S103, the difference Δd1^v=d1^v_max-d1^v_min is compared with a predefined threshold, according to the following relationship:

Δd1^v=d1^v_max−d1^v_min<Var_moy

Operations S104 or S105 of FIG. 4 take place in the same way as in the first embodiment.

Operations S100 to S103 and S104 or S105 are iterated for the pixels p2₁, p2₂, . . . , p2_N that are located in the same position (for example the second pixel of the first row from the left), in each of the projected depth maps D₁^v, D₂^v, D₃^v, D₄^v, and so on up to the pixels pQ₁, pQ₂, . . . , pQ_N that are located in the same position (for example the last pixel of the last row from the left), in each of the projected depth maps D₁^v, D₂^v, D₃^v, D₄^v.

The conditional modification S2 is implemented in the same way as in the first embodiment.

7.3 Third Embodiment

According to this third embodiment, the calculation of the confidence level takes place in the same way as in the first embodiment.

The calculation of the reliability information takes place in the same way as in the first embodiment up to and including S102. Operations S103 to S104 or S105 are replaced by the following:

In the case of the pixel p1₁ of the projected depth map D₁^v, the differences d1^v₁-d1^v_min and d1^v_max-d1^v₁ are each compared with a predefined threshold depth_TH, such that

$dept{h}_{TH}\frac{{\sum }_{j=1}^4\Delta d{1}_j}{4}.$

If d1^v₁-d1^v_min≥depth_TH, an item of reliability information F1₁ is set to a first value V3, such that for example V3=0, to indicate that the depth value of the pixel p1₁ in the projected depth map D₁^v is not reliable with respect to the minimum depth value d1^v_min.

As a variant, the comparison is d1^v₁-d1^v_min>depth_TH.

If d1^v₁-d1^v_min<depth_TH, an item of reliability information F1₁ is set to a second value V4, such that for example V4=1, to indicate that the depth value of the pixel p1₁ in the projected depth map D₁^v is reliable with respect to the minimum depth value d1^v_min.

As a variant, the comparison is d1^v₁-d1^v_min≤depth_TH.

If d1^v_max-d1^v₁≥depth_TH, an item of reliability information F1_1min is set to a first value V3, such that for example V3=0, to indicate that the depth value of the pixel p1₁ in the projected depth map D^v is not reliable with respect to the maximum depth value d1^v_max.

As a variant, the comparison is d1^v_max-d1^v₁>depth_TH.

If d1^v_max-d1^v₁<depthm, an item of reliability information F1_1max is set to a second value V4, such that for example V4=1, to indicate that the depth value of the pixel p1₁ in the projected depth map D₁^v is reliable with respect to the maximum depth value d1^v_max.

As a variant, the comparison is d1^v_max d1^v₁≤depth_TH.

The calculation described above is iterated for the pixel p1₂ of the projected depth map D₂^v, and so on up to the pixel p1₄ of the projected depth map D₄^v.

The calculation is then iterated for each following pixel p2₁ to pQ₁ of the projected depth map D₁^v, for each following pixel p2₂ to pQ₂ of the projected depth map D₂^v, for each following pixel p2₃ to pQ₃ of the projected depth map D₃^v, for each following pixel p2₄ to pQ₄ of the projected depth map D₄^v.

The conditional modification S2 is then implemented in the following manner.

For a given pixel pi₁ (1≤i≤Q) of the projected depth map D₁^v, it is determined whether:

• • the reliability information Fi_1min, associated with the pixel pi₁, is at 0 or 1,
  • the reliability information Fi_1max, associated with the pixel pi₁, is at 0 or 1.

If Fi_1min=0 and Fi_1max=0, the depth value di^v₁ of the pixel pi₁ of the projected depth map D₁^v is not modified.

If Fi_1min=1 and Fi_1max=0, the depth value di^v_1min of the pixel pi₁ of the projected depth map D₁^v is modified in S23 according to the following relationship:

dl_1min^v=mean((Ai₁, Ci₁, di^v₁) and/or (Ai₂, Ci₂, di^v₂) and/or (Ai₃, Ci₃, di^v₃) and/or (Ai₄, Ci₄, di^v₄)), depending on whether Ai₂=V1 or Ai₂=V2, Ai₃=b1 or Ai₃=V2, Ai₄=V1 or Ai₄=V₂.

According to one particular example, only the depth values di^v₁ and di^v₂ are taken into account in the modification of di^v_1min. The depth value di^v_1min of the pixel pi₁ of the projected depth map D₁^v is then modified according to the following relationship:

$\stackrel{\_}{d_{l_{1\min }}^v}=\frac{C{i}_1*{\mathrm{di}}_1^v+C{i}_2*{\mathrm{di}}_2^v}{C{i}_1+C{i}_2}$

If Fi_1min=0 and Fi_1max=1, the depth value di^v_1max of the pixel pi₁ of the projected depth map D₁^v is modified according to the following relationship:

dl_1max^v=mean((Ai₁, Ci₁, di^v₁) and/or (Ai₂, Ci₂, di^v₂) and/or (Ai₃, Ci₃, di^v₃) and/or (Ai₄, Ci₄, di^v₄)), depending on whether Ai₂=V1 or Ai₂=V2, Ai₃=V1 or Ai₃=V2, Ai₄=V1 or Ai₄=V2.

According to one particular example, only the depth values di^v₁, di^v₃ and di^v₄ are taken into account in the modification of di^v_1max. The depth value di^v_1max of the pixel pi₁ of the projected depth map D₁^v is then modified according to the following relationship:

$\stackrel{\_}{d_{l_{1\max }}^v}=\frac{C{i}_1*{\mathrm{di}}_1^v+C{i}_3*{\mathrm{di}}_3^v+C{i}_4*{\mathrm{di}}_4^v}{C{i}_1+C{i}_3+C{i}_4}$

If Fi_1min=1 and Fi_1max=1:

• • the depth value di^v_1min of the pixel pi₁ of the projected depth map D₁^v is then modified according to the following relationship:

$\stackrel{\_}{d_{l_{1\min }}^v}=\frac{{\sum }_{j=1}^4\left(C{i}_j*d{i}_j^v\right)}{{\sum }_{j=1}^4\left(C{i}_j\right)}$

• • the depth value di^v_1max of the pixel pi₁ of the projected depth map D₁^v is then modified according to the following relationship:

$\stackrel{\_}{d_{l_{1\max }}^v}=\frac{{\sum }_{j=1}^4\left(C{i}_j*d{i}_j^v\right)}{{\sum }_{j=1}^4\left(C{i}_j\right)}$

The conditional modification S2 described above is applied to the maximum and/or minimum depth value of each of the pixels p1_j, p2_j, . . . , pQ_j of the projected depth map D₁^v and is iterated for each pixel of each of the other projected depth maps D₂^v, D₃^v, D₄^v.

On completion of the conditional modification S2, the following are obtained, according to the value of the items of reliability information Fi_jmin and Fi_jmax:

• • an unmodified projected depth map D₁^v or a modified projected depth map D_1min^v and/or a modified projected depth map D_1max^v,
  • an unmodified projected depth map D₂^v or a modified projected depth map D_2min^v and/or a modified projected depth map D_2max^v,
  • an unmodified projected depth map D₃^v or a modified projected depth map D_3min^v and/or a modified projected depth map
  • an unmodified projected depth map D₄^v or a modified projected depth map D_4min^v and/or a modified projected depth map 8. Exemplary Implementation of Synthesis Device

FIG. 5 shows the simplified structure of a synthesis device SYNT designed to implement the synthesis method according to any one of the particular embodiments of the invention.

According to one particular embodiment of the invention, the actions performed by the synthesis method are implemented by computer program instructions. To that end, the synthesis device SYNT has the conventional architecture of a computer and comprises in particular a memory MEM, a processing unit UT, equipped for example with a processor PROC, and driven by the computer program PG stored in memory MEM. The computer program PG comprises instructions for implementing the actions of the synthesis method such as described above when the program is executed by the processor PROC.

On initialization, the code instructions of the computer program PG are for example loaded into a RAM memory (not shown), before being executed by the processor PROC. The processor PROC of the processing unit UT implements in particular the actions of the synthesis method described above, according to the instructions of the computer program PG.

In the case where the images I₁ to I_N have been decoded beforehand by a decoder DEC, according to a suitable decoding method:

• • the synthesis device SYNT is arranged at the output of the decoder DEC, as illustrated in FIG. 6A, or
  • the synthesis device SYNT forms part of the decoder DEC, as illustrated in FIG. 6B.

It goes without saying that the embodiments which have been described hereinabove have been given purely by way of wholly non-limiting indication, and that numerous modifications can be easily made by the person skilled in the art without however departing from the scope of the invention.

Claims

1. A method comprising: synthesizing an image of a view on the basis of images of N views, where N≥2, implemented by an image synthesis device, wherein the synthesizing comprises:projecting, to a position corresponding to the image of the view to be synthesized, N depth maps associated with the N views, respectively, andfor at least one given pixel of at least one projected depth map, for which a depth value has been associated on completion of the projecting, modifying said depth value of said at least one given pixel if an item of reliability information associated with said depth value is at a certain value, said modification using the depth value of a pixel whose position corresponds to that of said at least one given pixel, in at least one other projected depth map, which generates at least one modified projected depth map.

2. The method as claimed in claim 1, wherein said modification uses a weighting of the depth value of the pixel whose position corresponds to said at least one given pixel, in said at least one other projected depth map.

3. The method as claimed in claim 2, comprising calculating a confidence level of a pixel of a given depth map as a variation in the depth value of said pixel, said variation corresponding to a projection error in terms of allowed number of pixels.

4. The method as claimed in claim 3, wherein the confidence level of a pixel is weighted by a coding parameter of the image of the view with which the depth map that contains said pixel is associated.

5. The method as claimed in claim 1, wherein the modification of the depth value of the given pixel comprises replacing said depth value with a value that is calculated on the basis of said depth value and of the depth value of the pixel whose position corresponds to that of said at least one given pixel in said at least one other projected depth map, said depth values each being weighted by their respective confidence level.

6. The method as claimed in claim 1, wherein the item of reliability information is generated as follows: determining, in the set of N projected depth maps, for one and the same position as that of the given pixel, which pixel has the maximum depth value and which pixel has the minimum depth value,calculating the difference between the maximum and minimum depth values,comparing the calculated difference with a threshold, andgenerating an item of reliability information, the value of which is dependent on the result of the comparison.

7. The method as claimed in claim 1, wherein the item of reliability information is generated as follows: determining, in the set of N projected depth maps, for one and the same position as that of the given pixel, which pixel has the maximum depth value and which pixel has the minimum depth value,calculating a difference between the depth value of said given pixel and the determined minimum depth value,comparing the calculated difference with a threshold,generating an item of reliability information with respect to the determined minimum depth value, the value of which is dependent on the result of the comparison,calculating another difference between the determined maximum depth value and the depth value of said given pixel,comparing the other calculated difference with said threshold, andgenerating an item of reliability information with respect to the determined maximum depth value, the value of which is dependent on the result of the comparison.

8. The method as claimed in claim 6, wherein the threshold is equal to the mean of the N variations in the depth value of each pixel whose position corresponds to that of said given pixel in their respective depth map.

9. The method as claimed in claim 6, wherein the threshold is equal to the mean of the N variances in the depth value of each pixel whose position corresponds to that of said given pixel in their respective depth map.

10. A device for synthesizing an image of a view on the basis of images of N views where N≥2, said synthesis device being comprising a processor that is configured to implement: projecting, to a position corresponding to the image of the view to be synthesized, N depth maps associated with the N views, respectively, andfor at least one given pixel of at least one projected depth map, for which a depth value has been associated on completion of the projecting, modifying the depth value of said at least one given pixel if an item of reliability information associated with said depth value is at a certain value, said modification using the depth value of a pixel whose position corresponds to that of said at least one given pixel, in at least one other projected depth map, which generates at least one modified projected depth map.

11. The method as claimed in claim 1, wherein the method further comprises: decoding a data signal representative of a set of images of N coded views, where N≥2, comprising decoding decoding the images of the N coded views, producing a set of images of N decoded views,performing the synthesizing on the basis of said set of images of N decoded views.

12. The device as claimed in claim 10 wherein the processor is further configured to decode a data signal representative of a set of images of N coded views, where N≥2, wherein the processor is further configured to: decode the images of the N coded views, producing a set of images of N decoded views, andperform the synthesizing on the basis of said set of images of N decoded views.

13. (canceled)

14. A non-transitory computer-readable information medium comprising instructions of a computer program, which when executed by a processor of a synthesizing device configure the synthesizing device to synthethize an image of a view on the basis of images of N views, where N≥2, wherein the synthesizing comprises: projecting, to a position corresponding to the image of the view to be synthesized, N depth maps associated with the N views, respectively, andfor at least one given pixel of at least one projected depth map, for which a depth value has been associated on completion of the projecting, modifying said depth value of said at least one given pixel if an item of reliability information associated with said depth value is at a certain value, said modification using the depth value of a pixel whose position corresponds to that of said at least one given pixel, in at least one other projected depth map, which generates at least one modified projected depth map.

15. The method as claimed in claim 7, wherein the threshold is equal to the mean of the N variations in the depth value of each pixel whose position corresponds to that of said given pixel in their respective depth map.

16. The method as claimed in claim 7, wherein the threshold is equal to the mean of the N variances in the depth value of each pixel whose position corresponds to that of said given pixel in their respective depth map.

17. A device for decoding a data signal representative of a set of images of N coded views, where N≥2, said decoder comprising: a processor configured to decode the images of the N coded views, producing a set of images of N decoded views; anda device for synthesizing an image of a view on the basis of said set of images of N decoded views, said synthesis device comprising a processor that is configured to implement:projecting, to a position corresponding to the image of the view to be synthesized, N depth maps associated with the N decoded views, respectively, andfor at least one given pixel of at least one projected depth map, for which a depth value has been associated on completion of the projecting, modifying the depth value of said at least one given pixel if an item of reliability information associated with said depth value is at a certain value, said modification using the depth value of a pixel whose position corresponds to that of said at least one given pixel, in at least one other projected depth map, which generates at least one modified projected depth map.