CAMERA HAS TWO NATIVE IMAGES OUTPUT

Abstract

A new type of camera (100) comprises a pixel array (112), a filter array (114) optically coupled to the pixel array and a processing unit (120, 124) coupled to the pixel array. The filter array is formed by repeating a filter pattern (210) which comprises clear filters P1 in excess compared to infrared filters P2. On the one hand, the processing unit downsamples an image acquired by the pixel array so as to generate a first native output image (I1) formed of only the infrared pixels, an image transmitted to a facial recognition module. On the other hand, it applies a pixel defect correction mechanism to the acquired image so as to modify the minority infrared pixels into pixel values ​​homogeneous with the clear pixels, and thus generate a second native output image (I2) of the same dimensions as the acquired image. Figure for abstract: Figure 5

Description

Title of the invention: CAMERA WITH TWO NATIVE OUTPUT IMAGES Technical field

[0001] Embodiments and implementations relate to the field of signal processing and more particularly that of image sensors or “cameras”.

[0002] TECHNICAL CONTEXT

[0003] Conventional portable electronic equipment is generally equipped with a native digital camera, called a webcam, and an infrared (IR) camera. By "portable electronic equipment" we mean a portable computer (PC for "personal computer"), a tablet or a smartphone.

[0004] The webcam is typically used for acquiring images in the visible spectrum for digital applications (e.g., Internet communications). By "images" we mean both still images and images from a video sequence. Many different webcam models exist on the market, which evolve rapidly over time from one version of portable electronic equipment to another.

[0005] The infrared camera is traditionally combined with an infrared illumination device (usually near infrared or NIR) to capture images in the non-visible spectrum. The IR camera is generally used in the context of identification applications, such as facial recognition, where the use of infrared makes it possible to differentiate between a photograph or scan and a living person.

[0006] New applications based on the processing, by artificial intelligence (AI) engines, of images from the visible spectrum have emerged, such as computer vision.

[0007] Webcams appear to be poorly suited to these new applications due to their continuous evolution cycle. In particular, it may be necessary to relearn the AI ​​engine with each new generation of webcam, which necessarily induces additional costs (time and power consumption) in addition to the technical constraints of deploying the learned AI engines in the right equipment.

[0008] Furthermore, the classic presence of the webcam shutter can harm the effectiveness of these applications, because “misleading” images - because they are devoid of scene - are acquired and processed.

[0009] Adding a new (third) camera for these AI-based applications should be avoided both for cost reasons and for hardware integration and / or aesthetic reasons.

[0010] Also, there is a need to improve existing cameras.

[0011] SUMMARY

[0012] Taking up the above scenario, it is envisaged to develop the infrared camera into a camera with two native output images, capable of natively generating both an image in the visible spectral domain and an image in the IR domain. This improved IR camera has the advantage of evolving little from one generation of portable electronic equipment to another (unlike the webcam), while not being equipped with a shutter.

[0013] More generally, a camera with two native output images as described below can relate to other domains than infrared.

[0014] According to one aspect, there is therefore provided an image device comprising: a matrix of pixels (or photodiodes) for acquiring images, a matrix of pixel filters optically coupled to the matrix of pixels and formed by the repetition of a filter pattern, the filter pattern comprising pixel filters of a first type, called PI filters, in excess compared to one or more pixel filters of a second type, called P2 filters, of the filter pattern, a processing unit coupled to the matrix of pixels for (i) sub-sampling an image acquired by the matrix of pixels so as to generate a first output image formed of the pixels corresponding to the P2 filters, and for (ii) applying a pixel defect correction mechanism to the acquired image so as to generate a second output image of the same dimensions as the acquired image.

[0015] The sub-sampling of the so-called P2 pixels (optically corresponding to a P2 filter) makes it possible to obtain a first single-component image, with a resolution of 1 / Q (ratio of P2 filters in the repeated pattern), without any additional tool other than a conventional sub-sampling mechanism already existing.

[0016] Similarly, the application of the pixel defect correction mechanism (or "defect correction" in English) makes it possible, in an original way, to transform the P2 pixels of the acquired image into PI type pixels because they are extrapolated as a function of the neighboring PI pixels. And this, without any additional tool other than the classic defect correction mechanism already existing. A second single-component image is thus also obtained, with maximum resolution (that of the sensor pixel matrix).

[0017] Two high-quality single-component images are therefore generated during the same acquisition, without any additional tool to the mechanisms already existing in the sensors or image acquisition systems.

[0018] Such a camera with two native outputs can advantageously be used as IR camera (in combination with an NIR illumination device) for facial recognition identification operations and as a camera in the visible spectrum for AI applications, such as computer vision.

[0019] Also provided is an imaging system comprising a processing unit and an image device, as defined above, providing one of the first and second images to the processing unit when the latter is operating in a facial recognition identification mode and providing the other image to the processing unit when the latter is operating in a computer vision mode.

[0020] Correlatively, there is provided an image acquisition method comprising: receiving a signal from each pixel forming an image, of a pixel matrix optically coupled to a pixel filter matrix, the pixel filter matrix being formed by the repetition of a filter pattern, the filter pattern comprising pixel filters of a first type, called PI filters, in excess compared to one or more pixel filters of a second type, called P2 filters, of the filter pattern, sub-sampling the image pixel signal to generate a first output image formed from the pixels corresponding to the P2 filters, and applying a pixel defect correction mechanism to the image-forming pixel signal to generate a second output image of the same dimensions as said image.

[0021] The method has the same advantages as those of the aforementioned device.

[0022] Optional features of embodiments are defined in the appended claims. Some of these features are explained below with reference to a device, while they can be transposed into method features.

[0023] In one embodiment, the filter pattern is an n*m matrix filter comprising n*ml PI filters and a P2 filter (n and m are positive integers). The pattern is therefore made up of only PI and P2 filters. n=m for a square pattern; otherwise the pattern is rectangular. This configuration allows simple implementation and offers efficiency in pixel defect correction due to a large number of PI pixels neighboring each isolated P2 pixel to be “corrected”. The resulting quality of the second image is all the better as n and m are large.

[0024] In one embodiment, the filter pattern is one of:

[0025] (a) PI PI

[0026] PI P2

[0027] (b) PI PI PI

[0028] PI PI PI

[0029] PI PI P2

[0030] (c) PI PI PI

[0031] PI PI P2

[0032] (d) PI PI

[0033] PI PI

[0034] PI P2.

[0035] In one embodiment, any non-peripheral P2 filter in the filter array (i.e., not on the edges of the array) is surrounded by more PI filters than P2 filters. In other words, more than half of the neighboring filters of a P2 filter are PI filters. Filters at the edges of the array are exceptions in that they lack neighboring filters that would result from the pattern repetition.

[0036] The above arrangement ensures the isolation of the P2 pixels in the acquired image, for better detection of the P2 pixels as those to be "corrected", and therefore better efficiency of the pixel defect correction in order to generate the second image.

[0037] In one embodiment, any non-peripheral PI filter in the filter array is surrounded by as many or more PI filters as P2 filters. In other words, at least half of the neighboring filters of a PI filter are PL filters. This ensures that PI pixels will not be detected as different from neighboring pixels and therefore not detected as pixels to be "corrected". This contributes to obtaining a second quality image.

[0038] In certain embodiments, the P1 filters are clear filters (i.e. either without an optical filter or with a colorless optical filter) and the P2 filters are infrared filters (allowing all or part of the infrared spectrum to pass).

[0039] Alternatively, the P1 filters are infrared filters and the P2 filters are clear filters.

[0040] These configurations apply particularly well to the scenario discussed above. In particular, the image provided to the processing unit in a facial recognition identification mode is an infrared image, whereas the image provided to the processing unit in a computer vision mode is an image in the visible spectrum.

[0041] The device can effectively replace the IR camera of existing portable electronic equipment, for the purpose of providing an IR image for a facial recognition identification application and a visible spectrum image for an AI application, such as computer vision.

[0042] Of course, the clear filters can be replaced by filters on certain visible spectral sub-bands (for example blue, red, green). Similarly, configurations other than clear-IR can be envisaged, provided that substantially different, or even distinct, spectral bands are used for the PI and P2 filters.

[0043] In one embodiment, the fault correction mechanism is configured to apply a two-dimensional median filter (typically 3*3) to any pixel determined to be defective in the acquired image.

[0044] In particular, a pixel is determined as defective in the acquired image when its value is outside a range of values ​​depending on the values ​​taken by neighboring pixels. Typically, the limits of the range of values ​​are determined by the minimum and maximum values ​​taken by the neighboring pixels, possibly modified by a guard interval.

[0045] “Neighbor” means the pixels immediately adjacent to the pixel in question (first-rank pixels), or alternatively, also the second-rank pixels (i.e. neighbors of neighbors).

[0046] While it is of course possible to take into account all of the neighbors, certain embodiments favor the case where the neighboring pixels are neighboring pixels corresponding to PI filters. This arrangement allows better identification, by the defect correction mechanism, of only the P2 pixels.

[0047] Typically, the defect correction mechanism automatically identifies defective pixels and replaces them with the median filter. Of course, other corrective filters than the median filter can be used, for example a convolutional filter of the averaging type. Generally, the corrective filter used is based on the pixels neighboring the defective pixel. BRIEF DESCRIPTION OF THE FIGURES

[0048] Other advantages and characteristics of the invention will appear on examining the detailed description of the embodiment and implementation, which is in no way limiting, and the appended drawings in which:

[0049] [Fig.l] ;

[0050] [Fig.2] ;

[0051] [Fig.2A] ;

[0052] [Fig.3] ;

[0053] [Fig.4] ;

[0054] [Fig.5] ; and

[0055] [Fig.6] schematically illustrate methods of implementation and realization of the invention.

[0056] For the sake of clarity, the same elements are designated by the same references in the different figures. DETAILED DESCRIPTION

[0057] [Fig.l] illustrates, using a functional diagram, an imaging system 10 composed of an imaging device 100 with two native output images and an external processing unit 150, according to embodiments.

[0058] In an imaging device, a "native image" means an image that respects the proportions of the pixel sensor used and which results from a single specific type of filter used at the pixel sensor (i.e., a single-component image).

[0059] The two native images are either provided simultaneously as output to the external processing unit 150, or one or the other is provided to the unit 150 according to an active mode selected directly on the device or by downstream equipment.

[0060] The device 100 is for example a camera while the external processing unit 150 can be an image processing processor (or “ISP”) or any other processor in a computer embedding the camera 100. As a variant, the external processing unit 150 can be a separate, autonomous piece of equipment to which the external camera is connected (wired or not).

[0061] The imaging device 100 comprises an image sensor 110 and a controller 120 coupled to the sensor 110 and operating as a processing unit.

[0062] The image sensor 110 acquires images of a scene 199 by capturing incident light 198 and generates corresponding electrical or image signals, transmitted to the controller 120 for processing.

[0063] The image sensor 110 typically comprises a semiconductor substrate 111 provided with a plurality 112 of photodiodes 113 forming pixels, surmounted by a plurality 114 of filters 115 and a plurality of microlenses 116. The image sensor 110 may be arranged behind a set of lenses 130 forming the objective of the imaging device 100. These lenses may in particular be controlled by the controller 120 to adjust an optical power and carry out automatic focusing.

[0064] Each microlens 116 is optically coupled to a filter 115, itself optically coupled to a photodiode 113. The optical coupling can simply consist of a geometric alignment. Each photodiode 113 constitutes an elementary photosensitive sensor, for obtaining a pixel in an acquired image. Also, the photodiodes are hereinafter referred to as pixels.

[0065] [Fig.2] represents a matrix 200 of filtered pixels resulting from the superposition of a matrix 112 of photodiodes / pixels 113 with a matrix 114 of filters 115, optically coupled.

[0066] The filtered pixel matrix (and therefore each of the matrices 112 and 114) is a two-dimensional (2D) matrix organized into N columns CrCN of pixels P and M lines or rows Ri-Rm of pixels P with N and M natural integers. We denote i,j the column and line indices, so that each pixel is denoted Pij. Depending on the image format, N and M may be equal or different. A pixel matrix is ​​known to those skilled in the art; therefore, a more detailed description is not provided.

[0067] In the example of the Figure, the matrix 114 of filters 115 allowing this matrix of filtered pixels 200 to be obtained is formed by the repetition of a pattern 210 (in line bold) of filters 115. The pattern 210 is an n*m matrix filter, here with n=2 and m=2, of individual or elementary filters 115. Of course, other sizes of repeated pattern 210 can be used.

[0068] The pattern 210 comprises two types of filters, denoted PI (first type) and P2 (second type). The PI filters of the first type are in excess compared to the P2 filter(s) of the second type within the pattern 210. We denote 1 / Q the ratio of minority filters P2: 1 / Q = #P2 / n*m, where #P2 is the number of P2 filters in the pattern n*m.

[0069] The 2*2 pattern 210 of the Figure is for example a “3 for 1” pattern, namely three PI filters for a single P2 filter (1 / Q= 1 / 4). It is noted:

[0070] (a) PI PI

[0071] PI P2

[0072] More generally, an n*m “X for 1” pattern includes n*ml PI filters for a P2 filter (l / Q=l / (n*m)). For example, for n=m=3, the pattern 210 is as follows (l / Q=l / 9):

[0073] (b) PI PI PI

[0074] PI PI PI

[0075] PI PI P2

[0076] For n=3, m=2, the pattern 210 is as follows (1 / Q= 1 / 6):

[0077] (c) PI PI PI

[0078] PI PI P2

[0079] For n=2, m=3, the pattern 210 is as follows (1 / Q= 1 / 6):

[0080] (d) PI PI

[0081] PI PI

[0082] PI P2.

[0083] The position of the filter P2 in the pattern has little effect on the result of the processing described below, taking into account the repetition of the pattern in the matrix 114.

[0084] These different patterns 210, when repeated to form the matrix 114 of patterns, ensure that any P2 filter in the matrix 114, other than the filters at the periphery of the matrix, is surrounded by more PI filters than P2 filters, but also that any non-peripheral PI filter in the matrix 114 is surrounded by as many or more PI filters as P2 filters. Any pattern 210 more complex than those explained above which satisfies these conditions is suitable for implementation of the methods described below.

[0085] [Fig.2A] illustrates for example an alternative pattern 210' 2*3 comprising two neighboring P2 filters and four PI filters (l / Q=l / 3). Each PI filter has at most four neighboring P2 filters (and therefore at worst four neighboring PI filters) while each P2 filter has at most one neighboring P2 filter.

[0086] Returning to [Fig.l], the controller 120 controls the operation of the sensor image 110 and the lens set 130. It comprises one or more processors 121, a control unit 122, a reading unit 123, a processing unit 124 and an interface 125.

[0087] The processor 121 is typically coupled to random access and / or non-volatile memory (not shown) to execute instructions controlling the other units 122-125 of the controller 120 for implementing the methods described below.

[0088] The control unit 122 controls the operational characteristics of the lens set 130 and the pixel array 112, for example, the exposure time of the photodiodes, the time to acquire the pixel values ​​of the images.

[0089] The reading unit 123 reads or samples the analog signal coming from each of the individual photodiodes considered simultaneously or according to a predefined sequence, to generate a pixel signal, i.e. an image I with dimensions N*M. Such a reading unit 123 comprises in particular analog-digital converters and memory registers for the temporary storage of the acquired pixel values.

[0090] The processing unit 124 is coupled to the reading unit 123 to receive the pixel signal forming the image I of dimensions N*M and to process this image I according to the methods described below. The processing unit 124 can implement a large number of processing functionalities. For the present disclosure, the processing unit 124 includes a sub-unit 1240 for subsampling and a sub-unit 1242 for correcting pixel defects.

[0091] Subsampling is a function generally native to prior art devices / cameras. Subsampling techniques are well known to those skilled in the art; therefore, a detailed description of these techniques is not provided. As will be apparent from the remainder of the description, the subsampling subunit 1240 subsamples the acquired image I of dimensions N*M so as to generate a first output image h formed from the pixels corresponding to the minority filters P2.

[0092] Similarly, pixel defect correction is a function that is generally native to prior art devices / cameras. Pixel defect correction techniques are well known to those skilled in the art and are therefore not detailed here more than necessary. Generally speaking, the defect correction algorithms are implemented so as not to systematically discard sensors with defective pixels (photodiodes), and thus allow their use. As emerges from the remainder of the description, the sub-unit 1240 applies a pixel defect correction mechanism to the acquired image I of dimensions N*M so as to generate a second output image I2. This output image advantageously retains the maximum acquisition resolution N*M.

[0093] The interface 125 manages the exchange of input / output (I / O) information with the external processing unit 150.

[0094] [Fig. 3] illustrates, using a flowchart, steps of an image acquisition method 300. This method is notably carried out by the device 100 of [Fig. 1], and more precisely by the controller 120 using the sensor 110.

[0095] In the optional step 305 denoted S0, a pattern 210 of filters is chosen from a library of patterns, for example those presented above.

[0096] The choice of pattern influences the quality and speed of generation of the output image having the resolution N*M, as well as the resolution of the other generated image.

[0097] The choice of the pattern 210 can be carried out by an operator directly on the device / camera 100 (via for example a selection menu using the I / O interface 125) or by the equipment (computer) to which the device / camera 100 is connected, for example by the application executed on this equipment and which uses the images.

[0098] Step S0 thus makes it possible to configure, if necessary, the matrix 114 of filters 115, and in particular the repeated pattern 210. By way of example, all or part of the following parameters can be configured during this step: the dimensions n*m of the pattern, the ratio 1 / Q, the distribution of the filters PI and P2 in the pattern, the nature of the filters PI and P2.

[0099] In step 310 denoted SI, an image I of dimensions N*M is acquired using the sensor 110. The controller 120 receives a signal from each pixel of the matrix 112 of pixels 113 optically coupled to the matrix 114 of filters 115. As illustrated by [Fig.2] for example, this image I comprises pixels resulting from PI filtering and other pixels resulting from P2 filtering. This image is therefore not homogeneous at this stage, since it combines pixels of different natures.

[0100] Typically, the processing unit 124 receives the image I.

[0101] In step 315 denoted S2, the image I is processed for the sub-sampling sub-unit 1240. This sub-unit sub-samples the pixel signal I to generate a first output image I2 formed from the pixels corresponding to the filters P2. In practice, the sub-sampling simply preserves the pixels co-located with the filters P2. The size of the image I2 is then reduced by Q relative to the maximum resolution N*M: dimension(I2) = N*M / Q.

[0102] Advantageously, this operation S2 uses a sub-sampling function already present in the devices / cameras 100 of the state of the art.

[0103] In step 320 denoted S3, the image I is processed for the pixel defect correction sub-unit 1242.

[0104] Advantageously, the pixel defect correction algorithm applied to the image I automatically performs a detection of any defective pixels in the image and a correction thereof to generate a second output image I2 of the same dimensions as the image I.

[0105] The detection of defective pixels is carried out for example by determining the pixels whose value is outside a range of values ​​depending on the values ​​taken by neighboring pixels. The range of values ​​can be defined simply by the minimum and maximum values ​​of the neighboring pixels. Alternatively, this interval can be increased (in proportion to its extent) to avoid the correction of pixels whose value remains close to the neighboring pixels.

[0106] [Fig.4] illustrates an analyzed pixel Po surrounded by eight neighboring pixels Pi to P8, which are immediately adjacent. The minimum and maximum values ​​taken by pixels Pi to P8 are used to define a range of values ​​W, from which pixel Po is determined to be defective or not. Denoting p0 as the value of Po, if p0 does not belong to W, then Po is considered defective; if p0 belongs to W, then Po is considered not defective.

[0107] Due to the inhomogeneity of image I (arising from the two types of filters), this algorithm advantageously makes it possible to identify the minority type pixels (here type P2) in order to correct them according to the neighboring pixels, therefore of pixels predominantly of type PL II. This results in a correction homogenizing the image in the single component of type PL

[0108] Using pattern 210 of [Fig.2], the P2 type pixels (co-located with a P2 filter) are surrounded by PI type pixels only, leading to preferentially considering the P2 type pixels as being defective. Generally speaking, when the PI filters are in the majority, it is the P2 type pixels which are most likely to be detected as being defective. The conditions mentioned above on the ratio between the neighboring PI type pixels and those of the P2 type, lead to favoring the situation where the P2 type pixels are most likely to be detected as being defective.

[0109] Of course, other mechanisms for detecting defective pixels can be implemented. For example, the distance (difference) between p0 and the average of the values ​​of the neighboring pixels can be compared to a threshold value.

[0110] Some PI type pixels have a P2 type pixel as an immediate neighbor. Similarly, with certain patterns 210, a P2 type pixel may also have another P2 type pixel as an immediate neighbor. In one embodiment, provision may be made to consider as neighboring pixels for the detection of defective pixels only the PI type neighboring pixels, i.e. those collocated with PL filters. This configuration of course favors the detection of only P2 type pixels as being defective, for the purpose of correcting them.

[0111] If in the preceding example, the notion of neighborhood only retains the pixels PrP8 immediately adjacent to the pixel Po, variants may be required to consider the neighboring pixels of rank 2 (or even higher), that is to say the pixels neighboring the pixels im- mediately adjacent.

[0112] Each pixel detected as defective is then corrected by the defect correction algorithm of the subunit 1422. This makes it possible to convert the pixels from the P2 filters into pixel values ​​correlated with those of the PI type pixels. The objective of this operation is to obtain a single-component N*M full resolution image, corresponding to the component filtered by the PI type filters.

[0113] In one embodiment, the defect correction mechanism applies a two-dimensional median filter to any pixel determined to be defective in the I-image.

[0114] [Fig.4] illustrates the case of a 3*3 400 median filter. This modifies the value p0 into the median value p0' of the pixels PrP8. In this way, the new value p0' is homogeneous to the pixels obtained using the PL type filters. A full resolution image, PI filtered in all pixels is thus obtained while the matrix 114 of filters 115 is not entirely composed of PL filters.

[0115] In some embodiments, the applied filter 400 may exclude from the corrective calculation the defective neighboring pixels not yet corrected. This arrangement increases the quality of the image I2.

[0116] In certain embodiments, the filter 400 is applied by scanning the image according to a scanning pattern and therefore correcting the pixels progressively during the scanning of the image. This makes it possible to reduce the number of P2 type pixels taken into account in the detection and possible correction of subsequent pixels. The scanning pattern is advantageously chosen from:

[0117] a progressive scan line by line (or column by column),

[0118] a continuous progressive scan line by line (or column by column), that is to say without line return,

[0119] a diagonal zigzag sweep.

[0120] Of course, other defect correction techniques than the median filter can be used. The literature on the subject proposes a large number of them. For illustration purposes, an averaging filter, for example 3*3 (typically convolutional) can be used. The filter dimension 3*3 (media, averaging or other) is given here for illustration purposes only; any other dimension can be considered, for example 5*5. Furthermore, different weights can be applied to the values ​​of neighboring pixels.

[0121] Advantageously, operation S3 uses a fault correction function already present in the devices / cameras 100 of the state of the art.

[0122] Although represented successively, operations S2 and S3 can be carried out simultaneously, or in the reverse order to that described.

[0123] Steps S1 to S3 can be repeated for several successive images, in particular in the context of the acquisition of a video sequence.

[0124] [Fig.5] illustrates an application of the methods described above. In this ap plication, the device / camera 100 generates an image of the external scene 199 in the infrared spectrum for use in identifying an individual by facial recognition, and generates another image of the external scene 199 in the visible spectrum for computer vision use.

[0125] To do this, the PI (or P2) filters are clear filters (i.e. either without an optical filter or with a colorless optical filter) and the other filters are infrared filters.

[0126] The sensor 110 makes it possible, in collaboration with NIR illumination and the reading unit 123, to obtain the image I formed from a matrix of pixels filtered according to the repeated pattern 210, that is to say an inhomogeneous image with filtered pixels PI and filtered pixels P2.

[0127] In the example of the Figure, the light pixels correspond to the pixels obtained through a light filter, while the dark pixels correspond to the pixels obtained through an IR filter. The reverse is possible.

[0128] The sub-sampling 1240 of the minority type pixels (P2, infrared in the example) during step S2 makes it possible to obtain the image R formed of only the dark IR pixels. This image can then be provided as input to a facial recognition module FR.

[0129] The defect correction 1242 makes it possible, during step S3, to substitute the dark IR pixels with homogeneous values ​​for the light pixels and thus obtain a full resolution I2 image in the visible spectrum. This image can then be provided as input to a CV computer vision engine.

[0130] [Fig.6] illustrates a hardware architecture for the device 100 of [Fig.l]. It includes a communication bus 601 to which are preferably connected:

[0131] - one or more central processing units 602, such as one or more processors CPU and / or one or more microprocessors;

[0132] - a storage memory 603, of the ROM and / or flash memory type, for storage computer programs intended to implement all or part of the operations described above;

[0133] - a 604 RAM, of the RAM or video RAM (VRAM) type, for storage executable code of computer programs as well as the registers adapted to record variables and parameters necessary for their execution;

[0134] - a communication interface 605 connected to a network (not shown), a cable, bus or other means for exchanging information with external components or equipment (e.g. an ISP processor); and

[0135] - one or more inputs / outputs LO 606 allowing an operator to interact with computer programs, both in configuration and operation. Typically, inputs / outputs may include a screen serving as a graphical interface with the operator, which may be tactile or combined with a keyboard or other pointing device to allow the operator to interact with the programs.

[0136] The communication bus 601 ensures communication and interoperability between the different elements included in the computing device 100 or connected to it.

[0137] The central unit 602 is preferably adapted to control and direct the execution of the instructions or parts of software code of the computer program(s). When the power is switched on, the program(s) stored in non-volatile memory 603 are transferred / loaded into the RAM 604, which then contains the executable code of the program(s), as well as registers for storing the variables and parameters necessary for implementing the methods described.

[0138] Of course, the present disclosure is not limited to the embodiments described above as examples; it extends to other variants. Other embodiments are possible.

[0139] For example, if the subsampling 1420 and defect correction 1422 functions are illustrated inside the controller 120 and the device 100, they can be performed by an external component, placed before the facial recognition module FR and the computer vision engine CV in the example above.

Claims

Claims

1. An image device (100) comprising: an array (112) of pixels (113) for acquiring images (I), an array (114) of pixel filters (115) optically coupled to the array of pixels and formed by repeating a filter pattern (210), the filter pattern comprising pixel filters of a first type, called PI filters, in excess compared to one or more pixel filters of a second type, called P2 filters, of the filter pattern, a processing unit (120, 124) coupled to the array of pixels for (i) sub-sampling an image acquired by the array of pixels so as to generate a first output image (IJ) formed of the pixels corresponding to the P2 filters, and for (ii) applying a pixel defect correction mechanism to the acquired image so as to generate a second output image (I2) of the same dimensions as the acquired image.

2. The device of claim 1, wherein the filter pattern (210) is an n*m matrix filter comprising n*ml PI filters and a P2 filter.

3. A device according to claim 1 or 2, wherein the filter pattern (210) is one of: (a) PI PI PI P2 (b) PI PI PI PI PI PI PI PI P2 (c)Pl PI PI PI PI P2 (d) PI PI PI PI PI P2.

4. A device according to any preceding claim, wherein any non-peripheral P2 filter in the filter array (114) is surrounded by more PI filters than P2 filters.

5. A device according to any preceding claim, wherein any non-peripheral PI filter in the filter array (114) is surrounded by as many or more PI filters as P2 filters.

6. Device according to one of claims 1 to 5, in which the PI filters are clear filters and the P2 filters are infrared filters.

7. Device according to one of claims 1 to 5, in which the P1 filters are infrared filters and the P2 filters are clear filters.

8. Device according to one of the preceding claims, in which the defect correction mechanism is configured to apply a two-dimensional median filter (400) to any pixel (Pîj) determined to be defective in the acquired image (I).

9. Device according to claim 8, in which a pixel is determined as defective in the acquired image when its value (p0) is outside a range of values ​​(W) depending on the values ​​taken by neighboring pixels.

10. Device according to claim 9, in which the neighboring pixels are neighboring pixels corresponding to PI filters.

11. An imaging system (10) comprising a processing unit (150, CV, FR) and an image device (100) according to one of the preceding claims, providing one of the first and second images (Ib I2) to the processing unit when the latter is operating in a facial recognition identification mode and providing the other image to the processing unit when the latter is operating in a computer vision mode.

12. System according to claim 11, wherein the image provided (R) to the processing unit (150, FR) in a facial recognition identification mode is an infrared image, while the image (I2) provided to the processing unit (150, CV) in a computer vision mode is an image in the visible spectrum.

13. A method (300) for acquiring images comprising: receiving (310) a signal from each image pixel (I), of an array (112) of pixels (113) optically coupled to an array (114) of pixel filters (115), the array of pixel filters being formed by repeating a filter pattern (210), the filter pattern comprising pixel filters of a first type, called PI filters, in excess compared to one or more pixel filters of a second type, called P2 filters, of the filter pattern, sub-sampling (315) the image pixel signal to generate a first output image (h) formed of the pixels corresponding to the P2 filters, and applying (320) a pixel defect correction mechanism to the image pixel signal to generate a second output image (I2) of the same dimensions as said image.

Images