METHOD AND DEVICE FOR ESTIMATING A DEPTH MAP ASSOCIATED WITH A DIGITAL HOLOGRAM REPRESENTING A SCENE AND COMPUTER PROGRAM ASSOCIATED

Abstract

A method for estimating a depth map associated with a hologram representing a scene, the method includes steps of: reconstruction of images of the scene, each image being associated with a depth; decomposition of each image into a plurality of thumbnails adjacent to each other, each thumbnail being associated with the depth and including a plurality of pixels; determination, for each thumbnail, of a focus map by supplying, at the input of a network of neurons, values associated with the pixels of the thumbnail, to obtain, at the output of the network, the focus map including a focus level associated with the pixel concerned; and determination of a depth value, for each point of a depth map, as a function of the focus levels obtained. The invention also relates to an estimation device and an associated computer program.

Claims

1. Method for estimating a depth map associated with a digital hologram representing a scene, the method comprising: reconstructing a plurality of images of the scene from the digital hologram, each of the reconstructed images being associated with a depth of the scene, decomposing each reconstructed image into a plurality of thumbnails, said thumbnails of said plurality of thumbnails being adjacent to each other, each said thumbnail being associated with the depth of the scene corresponding to the reconstructed image concerned, each said thumbnail comprising a plurality of pixels, determining, for each said thumbnail, a focus map by supplying, at an input of an artificial neural network, values associated with said pixels of the thumbnail concerned, so as to obtain, at an output of the artificial neural network, said focus map comprising a focus level associated with the pixel concerned, and determining a depth value, for each point of the depth map, as a function of the focus levels obtained respectively for the pixels associated with said point of the depth map in the focus maps respectively determined for the thumbnails containing a given said pixel corresponding to said point of the depth map.

2. The method according to claim 1, wherein, for each said point of the depth map, the associated depth value is obtained by determining the depth corresponding to a highest said focus level among all the pixels associated with said point of the depth map in the thumbnails containing said pixels.

3. The method according to claim 1, wherein the artificial neural network is a convolutional neural network.

4. The method according to claim 1, wherein the reconstructing is implemented in such a way that the reconstructed images are respectively associated with said depths uniformly distributed between a minimum depth and maximum depth of the scene.

5. Device for estimating a depth map associated with a digital hologram representing a scene, the device comprising: a reconstruction module of a plurality of images of the scene from the digital hologram, each of the reconstructed images being associated with a depth of the scene, a decomposition module of each reconstructed image into a plurality of thumbnails, said thumbnails of said plurality of thumbnails being adjacent to each other, each said thumbnail being associated with the depth of the scene corresponding to the reconstructed image concerned, each said thumbnail comprising a plurality of pixels, a determination module, for each said thumbnail, a focus map by supplying, at an input of an artificial neural network, values associated with said pixels of the thumbnail concerned, so as to obtain, at an output of the artificial neural network, said focus map comprising a focus level associated with the pixel concerned, and a module for determining a depth value for each point of the depth map according to the focus levels obtained respectively for the pixels associated with said point of the depth map in the thumbnails containing said pixels.

6. A non-transitory computer-readable medium on which is stored a computer program comprising instructions executable by a processor and designed to implement the method according to claim 1 when the instructions are executed by the processor.

7. The method according to claim 2, wherein the artificial neural network is a convolutional neural network.

8. The method according to claim 2, wherein the reconstructing is implemented in such a way that the reconstructed images are respectively associated with said depths uniformly distributed between a minimum depth and maximum depth of the scene.

9. The method according to claim 3, wherein the reconstructing is implemented in such a way that the reconstructed images are respectively associated with said depths uniformly distributed between a minimum depth and maximum depth of the scene.

10. The method according to claim 7, wherein the reconstructing is implemented in such a way that the reconstructed images are respectively associated with said depths uniformly distributed between a minimum depth and maximum depth of the scene.

11. A non-transitory computer-readable medium on which is stored a computer program comprising instructions executable by a processor and designed to implement the method according to claim 2 when the instructions are executed by the processor.

12. A non-transitory computer-readable medium on which is stored a computer program comprising instructions executable by a processor and designed to implement the method according to claim 3 when the instructions are executed by the processor.

13. A non-transitory computer-readable medium on which is stored a computer program comprising instructions executable by a processor and designed to implement the method according to claim 4 when the instructions are executed by the processor.

14. A non-transitory computer-readable medium on which is stored a computer program comprising instructions executable by a processor and designed to implement the method according to claim 7 when the instructions are executed by the processor.

15. A non-transitory computer-readable medium on which is stored a computer program comprising instructions executable by a processor and designed to implement the method according to claim 8 when the instructions are executed by the processor.

16. A non-transitory computer-readable medium on which is stored a computer program comprising instructions executable by a processor and designed to implement the method according to claim 9 when the instructions are executed by the processor.

17. A non-transitory computer-readable medium on which is stored a computer program comprising instructions executable by a processor and designed to implement the method according to claim 10 when the instructions are executed by the processor.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0027] In addition, various other characteristics of the invention emerge from the appended description made with reference to the drawings which illustrate non-limiting forms of embodiment of the invention and where:

[0028] FIG. 1 represents, in a functional form, a device for estimating a depth map designed to implement a method for estimating a depth map in accordance with the invention,

[0029] FIG. 2 represents an example of a digital hologram associated with the depth map estimated according to the invention,

[0030] FIG. 3 is a schematic representation of an example of architecture of an artificial neural network (or a network of artificial neurons) implemented during the method of estimating a depth map according to the invention, and

[0031] FIG. 4 represents, in a flowchart form, an example of a method for estimating a depth map according to the invention.

[0032] It should be noted that, in these figures, the structural and/or functional elements common to the different variants may have the same references.

[0033] FIG. 1 represents, in a functional form, a device 1 for estimating (also denoted device 1 in the following) a depth map C from a digital hologram H.

[0034] The digital hologram H represents a given three-dimensional scene. This three-dimensional scene comprises, for example, one or more objects. The three-dimensional scene is defined in a marker (O, x, y, z).

[0035] As shown in FIG. 2, the digital hologram H is defined by a matrix of pixels in the (x, y) plane. The z axis, called the z depth axis, is orthogonal to the (x, y) plane of the digital hologram H. As can be seen in FIG. 2, the digital hologram H is here defined in the equation plane z=0.

[0036] The digital hologram H, for example, has a size of 1024?1024 pixels here.

[0037] The device 1 for estimating a depth map C is designed to estimate the depth map C associated with the digital hologram H. For this, the device 1 comprises a processor 2 and a storage device 4. The storage device 4 is for example a hard disk or a memory.

[0038] The device 1 also comprises a set of functional modules. It comprises for example a reconstruction module 5, a decomposition module 6, a module 8 for determining a focus map C.sub.i;j,k (or focusing map) and a module 9 for determining a depth value d.sub.js+q,ks+r.

[0039] Each one of the different modules described is for example implemented by means of computer program instructions designed to implement the module concerned when these instructions are executed by the processor 2 of the device 1 for estimating the depth map C.

[0040] However, as a variant, at least one of the aforementioned modules can be implemented by means of a dedicated electronic circuit, for example an integrated circuit with a specific application.

[0041] The processor 2 is also designed to implement an artificial neural network NN, involved in the process of estimating the depth map C associated with the digital hologram H.

[0042] An example of architecture of this artificial neural network NN is shown in FIG. 3. In this example, the artificial neural network NN is a convolutional neural network, for example of the U-Net type.

[0043] Generally speaking, such an artificial neural network NN comprises a plurality of convolution layers distributed according to different levels, as explained below and represented in FIG. 3. More details on an artificial neural network of the U-Net type can also be found in the article U-Net: Convolutional Networks for Biomedical Image Segmentation by Ronneberger, O., Fischer, P. & Brox, T., CoRR, abs/1505.04597, 2015.

[0044] In order to describe the architecture of the artificial neural network NN, we consider here that an image I.sub.e is provided at an input of this network of artificial neurons NN. In practice, this image I.sub.e is an image derived from the digital hologram H, as will be explained subsequently.

[0045] As shown in FIG. 3, the artificial neural network NN here comprises a first part 10, a connecting bridge 20 and a second part 30.

[0046] The first part 10 is a so-called contraction part. Generally speaking, this first part 10 has the encoder function and makes it possible to reduce the size of the image provided at an input while retaining (saving) its characteristics. For this, it comprises four levels here 12, 14, 16, 18. Each level 12, 14, 16, 18 comprises a convolution block Cony and a subsampling block D.

[0047] The convolution block Cony comprises at least one convolution layer whose kernel is a matrix of size n?n. Preferably here, each convolution block has two successive convolution layers. Here, each convolution layer has a kernel with a matrix of size 3?3.

[0048] Then, the convolution layer (or convolution layers if there are several) is followed by an activation function of rectified linear unit type (or ReLu for Rectified Linear Unit according to the commonly used designation of Anglo-Saxon origin). Finally, the convolution block Cony comprises, for the result obtained after application of the activation function, a so-called batch normalization. Here, this batch is composed by (or constituted by) the number of images provided as an input to the artificial neural network NN. During the learning step of the artificial neural network as described below, the batch size is greater than or equal to 1 (that means. at least two images are provided at the input to allow training the network of artificial so neurons). In the case of the depth map estimation method as described subsequently, the batch size is, for example here, given by the number of reconstructed images I.sub.i (see below).

[0049] As shown in FIG. 3, at the output of the convolution block Cony, each level 12, 14, 16, 18 comprises the subsampling block D. This subsampling block D makes it possible to reduce the dimensions of the result obtained at the output of the convolution block Cony. This involves, for example, a reduction by 2 of these dimensions, for example by selecting the maximum pixel value among the four pixels of a pixel window of size 2?2 (we then speak of max pooling 2?2 according to the commonly used Anglo-Saxon expression).

[0050] Thus, taking the example shown in FIG. 3, the input image I.sub.e is provided at an input to the first level 12 of the first part 10. The convolution block Cony and the subsampling block D of this first level 12 then make it possible to obtain, at the output, a first data X.sup.0, 0. Here, this first data X.sup.0, 0 has for example dimensions reduced by half compared to the input image I.sub.e.

[0051] Then, this first data X.sup.0, 0 is provided as an input to the second level 14 of the first part 10 so as to obtain, at the output thereof, a second data X.sup.1, 0. Here, this second data X.sup.1, 0 has for example dimensions reduced by half compared to the first data X.sup.0, 0.

[0052] The second data data X.sup.1, 0 is provided as an input to the third level 16 of the first part 10 so as to obtain, at the output thereof, a third data X.sup.2, 0. Here, this third data X.sup.2, 0 has for example dimensions reduced by half compared to the second data X.sup.1, 0.

[0053] Then, this third data X.sup.2, 0 is provided as an input to the fourth level 18 of the first part 10 so as to obtain, at the output, a fourth data X.sup.3, 0. Here, this fourth data X.sup.3, 0 has for example dimensions reduced by half compared to the third data X.sup.2, 0.

[0054] Thus, the processing operations of the input image I.sub.e by the first part 10 of the artificial neural network NN can be expressed in the following form:

X.sup.i,j=D(Conv(X.sup.i-1,j)) [0055] with j=0 and i between 1 and 4, the operator Cony corresponding to the processing implemented by the convolution block Cony and the operator D being associated with the subsampling block D.

[0056] As can be seen in FIG. 3, the artificial neural network NN comprises, at the output of the first part 10, the connection bridge 20. This connection bridge 20 makes it possible to make the link between the first part 10 and the second part 30 of the artificial neural network NN. It comprises a convolution block Cony as described previously. Here, it thus receives as an input the fourth data X.sup.3, 0 and provides, as an output, a fifth data X.sup.4, 0.

[0057] The second part 30 of the artificial neural network NN is called expansion. Generally speaking, this second part 30 has the decoder function and makes it possible to form an image having the size of the image provided at the input and which only contains the characteristics essential to the processing.

[0058] For this, the second part 30 here comprises four levels 32, 34, 36, 38. By analogy with the first part 10, we define the first level 38 of the second part 30 as that positioned at the same level as the first level 12 of the first part 10. The second level 36 of the second part 30 is positioned at the same level as the second level 14 of the first part 10 of the artificial neural network NN. The third level 34 of the second part is positioned at the same level as the third level 16 of the first part 10 of the artificial neural network NN. Finally, the fourth level 32 of the second part 30 is positioned at the same level as the fourth level 18 of the first part 10 of the artificial neural network NN. This definition is used to match the levels of the artificial neural network processing data of the same dimensions.

[0059] Each level 32, 34, 36, 38 comprises an oversampling block U, a concatenation block Conc and a convolution block Cony (such as that introduced previously in the first part).

[0060] Each oversampling block U aims at increasing the dimensions of the data received at an input. This is an upscaling operation according to the commonly used Anglo-Saxon expression. For example here, the dimensions are multiplied by 2.

[0061] Following the oversampling block U, each level 32, 34, 36, 38 comprises the concatenation block Conc. The latter aims at concatenating the data obtained at the output of the oversampling block U of the level concerned with the data of the same size obtained at the output of one of the levels 12, 14, 16, 18 of the first part 10 of the artificial neural network NN. The involvement of data from the first part of the artificial neural network NN in the concatenation operation is shown in broken lines in FIG. 3.

[0062] This concatenation block then allows the transmission of information of the extracted high frequencies obtained in the first part 10 of the artificial neural network NN also in the second part 30. Without this concatenation block Conc, this information could be lost following the multiple operations of undersampling and oversampling present in the artificial neural network NN.

[0063] Then, at the output of the concatenation block Conc, each level 32, 34, 36, 38 of the second part 30 comprises a convolution block Cony such as that described previously in the first part 10 of the artificial neural network NN. Here, each convolution block Cony notably comprises at least one convolution layer followed by a rectified linear unit type activation function and a batch normalization operation.

[0064] Based on the example shown in FIG. 3, the fifth data X.sup.4, 0 is provided at the input of the fourth level 32 of the second part 30. The oversampling bloc U then makes it possible to obtain at the output a first intermediate data X.sup.int1, which has the same dimensions as the fourth data X.sup.3, 0 obtained at the output of the fourth level 18 of the first part 10. This first intermediate data X.sup.int1 and the fourth data X.sup.3, 0 then are concatenated by the concatenation block Conc. The result obtained at the output of the concatenation block Conc is then provided as an input to the convolution block Cony so as to obtain, at the output, a sixth data item X.sup.3, 1.

[0065] That sixth data item X.sup.3, 1 then is provided at an input of the third level 34 of the second part 30 and, especially, at an input of the oversampling bloc U. At the output of that oversampling bloc U, a second intermediate data X.sup.int2, which has the same dimensions as the third data X.sup.2, 0, is obtained. The second intermediate data X.sup.int2 and the third data X.sup.2, 0 are concatenated by the concatenation block Conc. The result obtained at the output of the concatenation block Conc is provided at the input of the convolution block Conv so as to obtain, at the output, a seventh data item X.sup.2, 2.

[0066] Then, as shown in FIG. 3, the seventh data X.sup.2, 2 is provided at an input of the second level 36 of the second part 30 (and therefore at an input of the oversampling block U of this second level 36). A third intermediate data X.sup.int3 is obtained at the output of this oversampling block U. This third intermediate data X.sup.int3 has the same dimensions as the second data X.sup.1, 0. The third intermediate data X.sup.int3 and the second data X.sup.1, 0 are then concatenated by the concatenation block Conc. The result obtained at the output of the concatenation block Conc is provided at an input of the convolution block Conv so as to obtain, at the output, an eighth data item X.sup.1, 3.

[0067] Then, this eighth data X.sup.1, 3 is provided at an input of the first level 38 of the second part 30. The oversampling block U then makes it possible to obtain a fourth data X.sup.int4. The latter has the same dimensions as the first data X.sup.0, 0. The fourth intermediate data X.sup.int4 and the first data X.sup.0, 0 are then concatenated by the concatenation block Conc. The result obtained at the output of the concatenation block Conc is provided at an input of the convolution block Conv so as to obtain, at the output, a final data X.sup.0.4. This final data X.sup.0.4 has the same dimensions and the same resolution as the input image I.sub.e. In practice, here, this final data X.sup.0, 4 is for example associated with a focus map (also denoted focusing map) as described below.

[0068] Thus, the processing operations of the fifth data item X.sup.4, 0 by the second part 30 of the artificial neural network NN can be expressed in the following form:

X.sub.i,j=Conv(Conc[X.sup.i,0;U(X.sup.i+1,j?1)]) [0069] with j?1 and i between 0 and 3, the operator Conv corresponding to the processing implemented by the convolution block Conv, the operator Conc corresponding to the processing implemented by the concatenation block Conc and the operator U being associated with the oversampling block U.

[0070] FIG. 4 is a flowchart representing an example of a method (or a process) for estimating the depth map C associated with the digital hologram H, implemented in the context described above. This method is for example implemented by the processor 2. Generally, this process is implemented by computer.

[0071] As shown in FIG. 4, the method begins at step E2 during which the processor 2 determines a minimum depth z.sub.min and a maximum depth z.sub.max of the z coordinate in the three-dimensional scene of the digital hologram H. These minimum and maximum depths are for example previously recorded in the storage device 4.

[0072] The method then continues with a step E4 of reconstructing a plurality of two-dimensional images of the three-dimensional scene represented by the digital hologram H.

[0073] For this, the reconstruction module 5 is configured to reconstruct n images I.sub.i of the scene by means of the digital hologram H, with i being an integer ranging from 1 to n.

[0074] Each reconstructed image I.sub.i is defined in a reconstruction plane which is perpendicular to the depth axis of the digital hologram H. In other words, each reconstruction plane is perpendicular to the depth axis z. Each reconstruction plane is associated with a depth value, making it possible to associate a depth z.sub.i with each reconstructed image I.sub.i, the index i referring to the index of the reconstructed image I.sub.i. Each depth value defines a distance between the plane of the digital hologram and the reconstruction plane concerned.

[0075] Preferably here, the reconstruction step E4 is implemented in such a way that the depths z.sub.i associated with the reconstructed images I.sub.i are uniformly distributed between the minimum depth z.sub.min and the maximum depth z.sub.max. In other words, the reconstructed images I.sub.i are uniformly distributed along the depth axis, between the minimum depth z.sub.min and the maximum depth z.sub.max. Thus, the first reconstructed image I.sub.1 is spaced from the plane of the digital hologram H by the minimum depth z.sub.min while the last reconstructed image I.sub.n is spaced from the plane of the digital hologram H by the maximum depth z.sub.max.

[0076] The reconstruction planes associated with the reconstructed images I.sub.i are for example spaced two by two by a distance z.sub.e. The distance z.sub.e between each reconstruction plane is for example of the order of 50 micrometers (?m).

[0077] Preferably, the n images obtained in reconstruction step E4 are calculated using a propagation of the angular spectrum defined by the following formula:

[00001]$I_i\left(x,y\right)=F^{-1}\left\{F\left(H\right)e^{j2?z_i\sqrt{{?}^{-2}-f_{?}^2-f_y^2}}\right\}\left(x,y\right)$

with F and F.sup.?1 corresponding to direct and inverse Fourier transforms, respectively, and f.sub.x and f.sub.y being the frequency coordinates of the digital hologram H in the Fourier domain in a first spatial direction x and in a second spatial direction y of the digital hologram, ? being the acquisition wavelength of the digital hologram H, i being the index of the reconstructed image I with i ranging from 1 to n and z.sub.i being the depth given in the reconstruction plane of the image I.sub.i.

[0078] Each reconstructed image I.sub.i is defined by a plurality of pixels. Preferably, the reconstructed images are formed of as many pixels as the digital hologram H. Thus, the reconstructed images I.sub.i and the digital hologram H are of the same size. For example, in the case of a digital hologram H of size 1024?1024, each reconstructed image I.sub.i also has a size of 1024?1024.

[0079] As shown in FIG. 4, the method continues in step E6. During this step, the decomposition module 6 is configured to decompose each reconstructed image I.sub.i obtained in step E4 into a plurality of thumbnails J.sub.i; j,k. In other words, during this decomposition step E6, each reconstructed image I.sub.i is divided into a plurality of thumbnails J.sub.i;j,k. In other words still, each thumbnail J.sub.i;j,k corresponds to a sub-part of the reconstructed image I concerned.

[0080] Each thumbnail J.sub.i;j,k is defined by the following formula:

J.sub.i;j,k={|I.sub.i|(j.Math.s:(j+1).Math.s,k.Math.s:(k+1).Math.s)}

with

[00002]$j=1.Math..Math.\frac{s_W}{s}.Math.\mathrm{and}k=1.Math..Math.\frac{s_H}{s}.Math.,$

with s.sub.W and s.sub.H being the dimensions (respectively height and width) of the reconstructed image I.sub.i, s being the size of the thumbnail J.sub.i;j,k, |x.sub.1| being the notation corresponding to the module of the data x.sub.1 and ?x.sub.2?, the notation corresponding to the lower integer part of the number x.sub.2. The notation y.sub.1:y.sub.2 means that, for the variable concerned, the thumbnail J.sub.i; j,k is defined between pixel y.sub.1 and pixel y.sub.2. In other words, here, the previous formula defines the thumbnail J.sub.i; j,k, according to dimension x, between pixels js and (j+1)s of the reconstructed image I.sub.i and, according to dimension y, between pixels ks and (k+1)s of the reconstructed image I.sub.i.

[0081] Each thumbnail J.sub.i; j,k comprises a plurality of pixels. This plurality of pixels corresponds to a part of the pixels of the associated reconstructed image I.sub.i.

[0082] Here, the thumbnails J.sub.i; j,k are adjacent to each other. In practice, each thumbnail J.sub.i;j,k is formed from a set of contiguous pixels of the reconstructed image I.sub.i. Here, the sets of pixels of the reconstructed image I.sub.i (respectively forming each one of the thumbnails J.sub.i;j,k) are disjoint. In other words, this means that the thumbnails J.sub.i; j,k associated with a reconstructed image I.sub.i do not overlap with each other. Each thumbnail J.sub.i; j,k therefore comprises pixels which do not belong to the other thumbnails associated with the same reconstructed image I.sub.i. In other words, the thumbnails J.sub.i; j,k associated with a reconstructed image I.sub.i are independent of each other.

[0083] This property of independence between the thumbnails is particularly advantageous for the method according to the invention because it allows faster implementation. In addition, the necessary computing resources are less expensive thanks to the limited number of areas to analyze (namely the different thumbnails).

[0084] Since each thumbnail J.sub.i; j,k is derived from a reconstructed image I.sub.i associated with a depth z.sub.i, each thumbnail J.sub.i;j,k is also associated with this same depth z.sub.i (of the three-dimensional scene).

[0085] In the case where the digital hologram H has a size of 1024?1024, each thumbnail J.sub.i;j,k can for example have a size of 32?32.

[0086] In the case where the digital hologram H has a size of I.sub.H?I.sub.W, each thumbnail J.sub.i; j,k has a size of (32?s.sub.H)?(32?s.sub.W) with s.sub.H=I.sub.H/1024 and s.sub.W=I.sub.W/1024.

[0087] This definition of the size of the thumbnails makes it possible to ensure a size of these thumbnails adapted to the size of the digital hologram H so as to improve the speed of implementation of the method for estimating the depth map associated with the digital hologram H.

[0088] As shown in FIG. 4, the method then continues in step E8. During this step, the processor 2 determines, for each thumbnail J.sub.i; j,k, a focus map C.sub.i; j,k (or focusing map). This focus map C.sub.i; j,k includes a plurality of elements (each identified by the indices js+q, ks+r). Each element of the focusing map C.sub.i;j,k is associated with a pixel of the thumbnail J.sub.i; j,k concerned.

[0089] Here, each element of the focus map C.sub.i; j,k corresponds to a focus level (corresponding to a focus level associated with the pixel concerned in the thumbnail J.sub.i; j,k). In other words, the focusing map C.sub.i; j,k associates with each pixel of the thumbnail J.sub.i;j,k concerned a level of focus.

[0090] In practice, this step E8 is implemented via the artificial neural network NN., At an input, the latter receives each one of the thumbnails J.sub.i; j,k and provides at the output the focus levels (also denoted focusing levels) associated with each one of the pixels in the thumbnail J.sub.i; j,k concerned.

[0091] More particularly, the artificial neural network NN receives, at the input, each one of the pixels of the thumbnail J.sub.i; j,k and provides, at the output, the associated focus level (or focusing level). This focusing level is for example comprised between 0 and 1 and is equivalent to a level of sharpness associated with the pixel concerned. For example, in the case of a blurry pixel, the focus level is close to 0 while in the case of a noticeably sharp pixel, the focus level is close to 1.

[0092] Advantageously, the use of the artificial neural network allows faster processing of all the thumbnails and a more precise determination of the focusing levels associated with the pixels of the thumbnails.

[0093] Prior to implementing the estimation method, a learning step (not shown in the figures) allows the training of the artificial neural network NN. For this, computer-calculated holograms are used, for example. For these computed holograms, the exact geometry of the scene (and therefore the associated depth map) is known. A set of basic images comes from these calculated holograms.

[0094] For each base image in this set, each pixel is associated with a focus level. Indeed, for each pixel of each base image, the focus level is equal to 1 if the corresponding pixel, in the depth map, is equal to the associated depth. Otherwise, the focus level is 0.

[0095] The training step then consists of adjusting the weights of the nodes of the different convolution layers comprised in the different convolution blocks described previously so as to minimize the error between the focusing levels obtained at the output of the artificial neural network NN (when the basic images are provided at an input of this network) and those determined from the known depth map. For example, a crossed-entropy loss method can be used here in order to minimize the distance between the focus levels obtained at the output of the artificial neural network NN (when the base images are provided at an input of this network) and those determined from the known depth map.

[0096] In other words, the weights of the nodes of the different convolution layers are adjusted so as to converge the focus levels obtained at the output of the artificial neural network NN towards the focus levels determined from the known depth map.

[0097] In practice, the artificial neural network NN receives at the input all the thumbnails J.sub.i;j,k associated with each reconstructed image I.sub.i and proceeds to parallel processing of each of the thumbnails J.sub.i;j,k.

[0098] Alternatively, the thumbnails J.sub.i; j,k could be processed successively, one after the other.

[0099] At the end of step E8, the processor 2 therefore knows, for each thumbnail J.sub.i;j,k, the associated focusing map C.sub.i; j,k which lists the focusing levels obtained at the output of the artificial neural network NN associated with each pixel of the thumbnail J.sub.i; j,k concerned. Each focusing map C.sub.i; j,k is associated with the corresponding thumbnail J.sub.i;j,k, and thus with the depth z.sub.i (of the three-dimensional scene).

[0100] As shown in FIG. 4, the method then comprises a step E10 of estimating the depth map C associated with the digital hologram H. This depth map C comprises a plurality of depth values d.sub.js+q, ks+r. Each depth value d.sub.js+q, ks+r is associated with a pixel among the different pixels of the thumbnails J.sub.i; j,k. The depth value d.sub.js+q, ks+r is determined based on the focus levels determined in step E8.

[0101] Indeed, during step E8, as each reconstructed images of the plurality of reconstructed images I.sub.i corresponds to a different depth z.sub.i, a pixel of a thumbnail is associated with different focusing levels (depending on the depth of the reconstructed image I from which the thumbnail concerned is derived). In other words, for each pixel (associated with a depth value d.sub.js+q, ks+r of the depth map C), several focusing levels are known.

[0102] Thus, here, processor 2 determines, for each pixel associated with the depth value d.sub.js+q, ks+r concerned, the depth for which the focusing level is the highest:

d.sub.js+q,ks+r=argmax.sub.i=1 . . . N(C.sub.i;j,k(q,r)) [0103] with (js+q, ks+r) being the pixel to which the determined depth value is assigned, NN being the operator corresponding to the implementation of the artificial neural network NN and argmax, the operator translating the determination of the maximum value of the focus level (obtained at the output of the artificial neural network).

[0104] In other words, for each index pixel (js+q, ks+r), processor 2 determines the depth at which the focus level is highest. This depth then corresponds to the depth value d.sub.js+q, ks+r (element of depth map C).

[0105] Alternatively, the depth value could be determined using another method than determining the maximum value of the focus level. For example, an area formed by a plurality of adjacent pixels may be defined and the depth value may be determined by considering the depth for which a maximum deviation is observed from the average of the focus levels over the defined pixel area.

[0106] At the end of step E10, therefore, the depth map C is for example estimated here from these determined depth values. Thus, each element of the depth map C comprises a depth value d.sub.js+q, ks+r associated with each pixel having the index (js+q, ks+r).

[0107] This estimated depth map C ultimately makes it possible to have spatial information in the form of a matrix of depth values representing the three-dimensional scene associated with the digital hologram H.

[0108] Of course, the method described above for a digital hologram applies in the same way to a plurality of holograms. For a plurality of digital holograms, the implementation of the method can be successive for each hologram or in parallel for the plurality of digital holograms.