IMAGE CAPTURE USING RADIATION-SENSITIVE ELEMENTS HAVING A MEMORY EFFECT

Abstract

Disclosed is a method for capturing images that makes it possible to correct at least partially a memory effect of sensitive elements of a matrix used to capture the images. A corrected image is formed by subtracting, from a captured new raw image, a part of a prior raw image that was captured before the new raw image. The method is particularly suitable for sensitive elements with a first-order transfer function with respect to time, such as bolometers or microbolometers. Correction of the memory effect makes it possible to improve the transfer function and/or reduce a tail effect that is present in the images when scene elements move.

Claims

1. A method for capturing images, wherein several images are successively captured using a same matrix of sensitive elements, each sensitive element exhibiting a memory effect which makes a raw detection signal of said sensitive element depend on an amount of radiation received by said sensitive element at a read-out time at which the raw detection signal is read, but also depend on amounts of radiation received by said sensitive element before said read-out time, wherein, in order to form an image which is at least partially corrected for the memory effect, called corrected image, an image point intensity value is assigned separately to each of the sensitive elements of the matrix, said image point intensity value being proportional to a difference between the raw detection signal of the sensitive element as read for a new captured image, called new raw image, and a part of the raw detection signal of the same sensitive element as read for another image, called prior raw image, which was captured before said new raw image, the image point intensity value which is attributed to each of the sensitive elements in order to form the corrected image, being proportional to a result of dividing by [1-exp(-Δ/τ)] a difference between the raw detection signal of the sensitive element as read for the new raw image, and the result of multiplying by exp(-Δ/τ) the raw detection signal of the same sensitive element as read for the prior raw image, where exp(.) designates an exponential function, i is a characteristic response time of the sensitive element, and Δ is a non-zero duration between the read-out times of said sensitive element respectively for the new raw image and the prior raw image.

2. The method according to claim 1, wherein the duration A between the respective read-out times of a same sensitive element respectively for the new raw image and for the prior raw image, is less than the characteristic response time ti of said sensitive element.

3. The method according to claim , wherein the prior raw image the last one of several images that have been captured before the new raw image according to a chronological order of capture of the raw images.

4. The method according to claim 1, wherein each sensitive element is a bolometer or a microbolometer, a thermopile, a pyroelectric sensor, a ferroelectric sensor, or a thermally deformable microlever sensor.

5. The method according to claim 1, wherein each corrected image is formed from a pair of raw images, comprising a new raw image and a prior raw image that has been captured before said new raw image, and wherein the pairs of raw images used to form successive corrected images are disjoint, non-interlaced, and chronologically successive.

6. The method according to claim 1, wherein several corrected images are formed by combining a same new raw image with several prior raw images successively captured before the new raw image, and wherein each corrected image is obtained by using for the duration Δ relating to each sensitive element, the difference between the read-out times of said sensitive element respectively for said new raw image and for the prior raw image which are combined to form said corrected image.

7. An image sensor, comprising: a matrix of sensitive elements, each sensitive element exhibiting a memory effect which makes a raw detection signal that is read from said sensitive element depend on an amount of radiation received by said sensitive element at a read-out time at which the raw detection signal is read, but also depend on amounts of radiation received by said sensitive element before said read-out time, and - an image processing module, adapted for outputting images at least partially corrected of the memory effect, called corrected images, each corrected image being formed by image point intensity values which are respectively assigned to the sensitive elements of the matrix, the image processing module being adapted for calculating the image point intensity value of any one of the sensitive elements as being proportional to a difference between the raw detection signal of said sensitive element as read for a new captured image, called new raw image and a part of the raw detection signal of the same sensitive element as read for another image, called prior raw image, which was captured before said new raw image, the image point intensity value which is assigned to each of the sensitive elements in order to form the corrected image, being proportional to a result of dividing by [1 - exp(-Δ/τ)] a difference between the raw detection signal of the sensitive element as read for the new raw image and a result of multiplying by exp(-Δ/τ) the raw detection signal of the same sensitive element as read for the prior raw image, where exp(.) denotes an exponential function, i is a characteristic response time of the sensitive element, and Δ is a non-zero duration between the read-out times of said sensitive element respectively for the new raw image and the prior raw image.

8. An image sensor comprising: a matrix of sensitive elements, each sensitive element exhibiting a memory effect which makes a raw detection signal that is read from said sensitive element depend on an amount of radiation received by said sensitive element at a read-out time at which the raw detection signal is read, but also depend on amounts of radiation received by said sensitive element before said read-out time, and an image processing module, adapted for outputting images at least partially corrected of the memory effect, called corrected images, each corrected image being formed by image point intensity values which are respectively assigned to the sensitive elements of the matrix, the image processing module being adapted for calculating the image point intensity value of any one of the sensitive elements as being proportional to a difference between the raw detection signal of said sensitive element as read for a new captured image, called new raw image, and a part of the raw detection signal of the same sensitive element as read for another image, called prior raw image, which was captured before said new raw image, the image point intensity value which is assigned to each of the sensitive elements in order to form the corrected image, being proportional to a result of dividing by [1 - exp(-Δ/τ)] a difference between the raw detection signal of the sensitive element as read for the new raw image, and a result of multiplying by exp(-Δ/τ) the raw detection signal of the same sensitive element as read for the prior raw image, where exp(.) denotes an exponential function, t is a characteristic response time of the sensitive element, and Δ is a non-zero duration between the read-out times of said sensitive element respectively for the new raw image and the prior raw image, wherein the image sensor adapted for implementing the method of claim 1.

9. The method according to claim 2, wherein the prior raw image is the last one of several images that have been captured before the new raw image, according to a chronological order of capture of the raw images.

10. The method according to claim 2, wherein each sensitive element is a bolometer or a microbolometer, a thermopile, a pyroelectric sensor, a ferroelectric sensor, or a thermally deformable microlever sensor.

11. The method according to claim 3, wherein each sensitive element is a bolometer or a microbolometer, a thermopile, a pyroelectric sensor, a ferroelectric sensor, or a thermally deformable microlever sensor.

12. The method according to claim 2, wherein each corrected image is formed from a pair of raw images, comprising a new raw image and a prior raw image that has been captured before said new raw image, and wherein the pairs of raw images used to form successive corrected images are disjoint, non-interlaced, and chronologically successive.

13. The method according to claim 3, wherein each corrected image is formed from a pair of raw images, comprising a new raw image and a prior raw image that has been captured before said new raw image, and wherein the pairs of raw images used to form successive corrected images are disjoint, non-interlaced, and chronologically successive.

14. The method according to claim 4, wherein each corrected image is formed from a pair of raw images, comprising a new raw image and a prior raw image that has been captured before said new raw image, and wherein the pairs of raw images used to form successive corrected images are disjoint, non-interlaced, and chronologically successive.

15. The method according to claim 2, wherein several corrected images are formed by combining a same new raw image with several prior raw images successively captured before the new raw image, and wherein each corrected image is obtained by using for the duration A relating to each sensitive element, the difference between the read-out times of said sensitive element respectively for said new raw image and for the prior raw image which are combined to form said corrected image.

16. The method according to claim 3, wherein several corrected images are formed by combining a same new raw image with several prior raw images successively captured before the new raw image, and wherein each corrected image is obtained by using for the duration A relating to each sensitive element, the difference between the read-out times of said sensitive element respectively for said new raw image and for the prior raw image which are combined to form said corrected image.

17. The method according to claim 4, wherein several corrected images are formed by combining a same new raw image with several prior raw images successively captured before the new raw image, and wherein each corrected image is obtained by using for the duration A relating to each sensitive element, the difference between the read-out times of said sensitive element respectively for said new raw image and for the prior raw image which are combined to form said corrected image.

18. An image sensor comprising: a matrix of sensitive elements, each sensitive element exhibiting a memory effect which makes a raw detection signal that is read from said sensitive element depend on an amount of radiation received by said sensitive element at a read-out time at which the raw detection signal is read, but also depend on amounts of radiation received by said sensitive element before said read-out time, and an image processing module, adapted for outputting images at least partially corrected of the memory effect, called corrected images, each corrected image being formed by image point intensity values which are respectively assigned to the sensitive elements of the matrix, the image processing module being adapted for calculating the image point intensity value of any one of the sensitive elements as being proportional to a difference between the raw detection signal of said sensitive element as read for a new captured image, called new raw image, and a part of the raw detection signal of the same sensitive element as read for another image, called prior raw image, which was captured before said new raw image, the image point intensity value which is assigned to each of the sensitive elements in order to form the corrected image, being proportional to a result of dividing by [1 - exp(-Δ/τ)] a difference between the raw detection signal of the sensitive element as read for the new raw image, and a result of multiplying by exp(-Δ/τ) the raw detection signal of the same sensitive element as read for the prior raw image, where exp(.) denotes an exponential function, t is a characteristic response time of the sensitive element, and Δ is a non-zero duration between the read-out times of said sensitive element respectively for the new raw image and the prior raw image, wherein the image sensor adapted for implementing the method of claim 2.

19. An image sensor comprising: a matrix of sensitive elements, each sensitive element exhibiting a memory effect which makes a raw detection signal that is read from said sensitive element depend on an amount of radiation received by said sensitive element at a read-out time at which the raw detection signal is read, but also depend on amounts of radiation received by said sensitive element before said read-out time, and an image processing module, adapted for outputting images at least partially corrected of the memory effect, called corrected images, each corrected image being formed by image point intensity values which are respectively assigned to the sensitive elements of the matrix, the image processing module being adapted for calculating the image point intensity value of any one of the sensitive elements as being proportional to a difference between the raw detection signal of said sensitive element as read for a new captured image, called new raw image, and a part of the raw detection signal of the same sensitive element as read for another image, called prior raw image, which was captured before said new raw image, the image point intensity value which is assigned to each of the sensitive elements in order to form the corrected image, being proportional to a result of dividing by [1 - exp(-Δ/τ)] a difference between the raw detection signal of the sensitive element as read for the new raw image, and a result of multiplying by exp(-Δ/τ) the raw detection signal of the same sensitive element as read for the prior raw image, where exp(.) denotes an exponential function, t is a characteristic response time of the sensitive element, and Δ is a non-zero duration between the read-out times of said sensitive element respectively for the new raw image and the prior raw image, wherein the image sensor adapted for implementing the method of claim 3.

20. An image sensor comprising: a matrix of sensitive elements, each sensitive element exhibiting a memory effect which makes a raw detection signal that is read from said sensitive element depend on an amount of radiation received by said sensitive element at a read-out time at which the raw detection signal is read, but also depend on amounts of radiation received by said sensitive element before said read-out time, and an image processing module, adapted for outputting images at least partially corrected of the memory effect, called corrected images, each corrected image being formed by image point intensity values which are respectively assigned to the sensitive elements of the matrix, the image processing module being adapted for calculating the image point intensity value of any one of the sensitive elements as being proportional to a difference between the raw detection signal of said sensitive element as read for a new captured image, called new raw image, and a part of the raw detection signal of the same sensitive element as read for another image, called prior raw image, which was captured before said new raw image, the image point intensity value which is assigned to each of the sensitive elements in order to form the corrected image, being proportional to a result of dividing by [1 - exp(-Δ/τ)] a difference between the raw detection signal of the sensitive element as read for the new raw image, and a result of multiplying by exp(-Δ/τ) the raw detection signal of the same sensitive element as read for the prior raw image, where exp(.) denotes an exponential function, t is a characteristic response time of the sensitive element, and Δ is a non-zero duration between the read-out times of said sensitive element respectively for the new raw image and the prior raw image, wherein the image sensor adapted for implementing the method of claim 4.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The features and advantages of the invention will be more clearly apparent in the following detailed description of some examples of non-limiting implementations, with reference to the appended figures, in which:

[0032] [FIG. 1] is a block diagram of an image sensor according to the invention;

[0033] [FIG. 2] to [FIG. 4] illustrate different sequences for producing corrected images, in accordance with the invention;

[0034] [FIG. 5] illustrates a first advantage of the invention, concerning image contrast; and

[0035] [FIG. 6] illustrates a second advantage of the invention, concerning a tail effect possibly present in the images.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0036] According to [FIG. 1 1], an image sensor 10 comprises a matrix 10a of sensitive elements 1, which are arranged at the intersections of rows and columns of the matrix, and can be considered as independent of each other. For example, the matrix 10a may be composed of 320×240 sensitive elements 1. All the sensitive elements 1 may be identical, in particular of a model known to those skilled in the art. For example, each sensitive element 1 may be a microbolometer. It then comprises a portion of electrically conductive material whose electrical resistance value varies as a function of its temperature. This portion of variable electrical resistance material is at least partially thermally insulated from its environment, so that the radiation R that is received and absorbed by this portion causes an increase in its temperature, and consequently a variation in its electrical resistance value. This electrical resistance value constitutes the raw detection signal which is read for each captured image. Such measurement principle is very well known, so it is not necessary to describe it again here.

[0037] It is also known that a microbolometer of this type has a first-order transfer function with respect to time, which is characterized by a value of a gain coefficient denoted G, and by a value of a characteristic response time denoted τ. The transfer function, which is denoted by f(s) and is dependent on the Laplace variable s, is then

[00001]$f\left(s\right)=\frac{G}{1+\tau .Math.s}.$

In lighting conditions with an intensity of the radiation R which varies sinusoidally over time according to a frequency v, the detection signal produced by the sensitive element also varies over time at frequency v, with an amplitude which is given by the formula A.sub.d=f(j2π.Math.v).Math.A.sub.R, where A.sub.R is the complex amplitude of the intensity of the radiation R, A.sub.d is the complex amplitude of the raw detection signal, and j is the imaginary unit of the complex numbers. Commonly, the characteristic response time τ may be between 7 ms (millisecond) and 15 ms, and the value of the gain G depends in particular on the geometric and thermal features of the sensitive element. The detection signal produced by the sensitive element depends on the intensity of the radiation R which has been received at each instant until the moment when the detection signal is read. This behavior is called memory effect, and reduces the sensitivity of the sensitive element to rapid variations in the intensity of the radiation R. This reduction in sensitivity is due to an effect of weighted combination of the instantaneous values of the intensity of the radiation previously received by the sensitive element. This results in a temporal smoothing of these values in the detection signals produced by the sensitive element. The characteristic response time τ defines the time scale according to which a contribution occurs to the detection signal that is read, from radiation received at a time prior to that of the reading of the detection signal. Such contribution to the value of the detection signal that is read is affected by a multiplicative attenuation factor of the type exp(−t/τ), which applies to the intensity of the radiation received before the read-out time, where t is the duration between the time at which the radiation was received by the sensitive element and the read-out time.

[0038] To form the image sensor 10, the matrix 10a of sensitive elements 1 is associated with a controller 10b, denoted CTRL. In a known manner, the controller 10b has the power supply and reading functions for each of the sensitive elements 1, an addressing function for each sensitive elemeent 1 within the matrix 10a, and possibly additional functions such as a test function for the sensitive elements 1, an optional thermalization function for at least some of the sensitive elements, and a digitization of the detection signals. In each read-out cycle of all the sensitive elements 1 of the matrix 10a, the controller 10b outputs the values of the detection signals which have been read from all the sensitive elements 1, with one detection signal value read per sensitive element for each captured image. The read-out of these detection signals, which have been called raw detection signals in the general part of this description, constitutes an image capture operation, and the image thus directly obtained is called raw image. The read-out of the matrix of sensitive elements may be performed according to one of the following two modes: rolling shutter, or snapshot mode.

[0039] In the first case, the rows of the matrix are read sequentially, row by row, and in the second case, all rows are read at the same time. This difference in the reading mode of the image sensor in no way affects the principle of the invention, nor the results. The following detailed description is provided as a non-limiting example for the snapshot reading mode. To transpose it to the case of reading images using rolling shutter mode, one can consider for example the read-out times mentioned thereafter as being those of the first row of the matrix.

[0040] Although not shown in [FIG. 1 1], an optics may be used in front of the matrix 10a, in order to optically conjugate a scene to be imaged with the matrix 10a of the sensitive elements 1. The radiation R then passes through this conjugating lens before being incident on the sensitive elements 1.

[0041] For the invention, the image sensor 10 further comprises an image processing module 11, which is connected so as to receive at input the raw images outputted by the controller 10b. The module 11 is designed to produce processed images from the raw images, in order to compensate for or at least partially correct the memory effect of the sensitive elements 1 which has been described above. For this reason, the processed images produced according to the invention by the image sensor including this module 11 are called corrected images. The image processing module 11 may be a dedicated electronic circuit, or may be a software module hosted in a processor, denoted CPU.

[0042] According to the invention, the image processing module 11 produces a corrected image from a raw image, by subtracting from this raw image another raw image which was previously captured, multiplied by a determined coefficient. Multiplying a raw image by a coefficient is understood to mean the operation which consists of multiplying by this coefficient all the raw detection signals which constitute the raw image. Furthermore, subtraction of a first image from a second image is understood to mean the operation which consists, independently for each sensitive element 1, of calculating the difference between the detection signal which was read for the second image and the one read for the first image. Thus, according to the invention, a corrected image denoted S.sub.corr(t) may be obtained by calculating and grouping the values S.sub.i,j.sub.corr(t)=β.Math.[S.sub.i,j_raw(t)−α.Math.S.sub.i,j_raw(t-Δ)], where t is the read-out time of a new raw image, denoted S.sub.raw(t) and composed of raw detection signals S.sub.i,j_raw(t), t-Δ is the read-out time of a prior raw image, denoted S.sub.raw(t-Δ) and composed of raw detection signals S.sub.i,j_raw(t-Δ), A is the duration between the read-out times of both raw images S.sub.raw(t) and S.sub.raw(t-Δ), and S.sub.i,j_raw(t) is the raw detection signal as read at time t for sensitive element i,j in the matrix 10a, i and j respectively being the row and column numbers of the sensitive element considered. α is the multiplicative coefficient which is applied to the prior raw image S.sub.raw(t-Δ), and β is a global multiplicative coefficient. α and β are positive and non-zero. S.sub.i,j_corr(t) is then the image point intensity value of the corrected image S.sub.corr(t), for the image point associated with the sensitive element located at the intersection of the i.sup.th row and the j.sup.th column of the matrix 10a.

[0043] Advantageously, the coefficient a, and possibly also the coefficient β, may be selected to vary as a function of the duration between the read-out times of both raw images S.sub.raw(t-Δ) and S.sub.raw(t), which are separately composed of the read-out raw signals S.sub.i,j_raw(t-Δ) and S.sub.i,j_raw(t).

[0044] Possibly, but optionally, at least one of the coefficients α and β may have values which are different for sensitive elements 1 which are distinct in the matrix 10a. In this case, the values of the coefficients α and β may be determined separately for each sensitive element 1, during a calibration or benchmarking step which may be carried out before each image capture sequence, or in laboratory.

[0045] In preferred implementations of the invention, coefficient a may be determined in accordance with the equation:

[00002]$\alpha =\exp \left(-\frac{\Delta }{\tau }\right),$

where Δ again designates the duration between the respective read-out times of the sensitive element for both raw images S.sub.raw(t-Δ) and S.sub.raw(t), and τ again designates the characteristic response time of the sensitive element. Optionally, the value used for the characteristic response time T may vary according to the sensitive elemeent 1 inside the matrix 10a. In this case, the values of the characteristic response time τ for all sensitive elements 1 may have been determined separately for each sensitive element during the calibration or benchmarking step, carried out before each image capture sequence or in laboratory. They are then stored within the image processing unit 11 or in a memory which is accessible to the unit 11.

[0046] In even more preferred implementations of the invention, coefficient β may be determined according to the equation:

[00003]$\beta =\frac{a}{1-\exp \left(-\frac{\Delta }{\tau }\right)},$

where a is a non-zero constant which can set a scale for the image point intensity values of the corrected images. For these implementations, the image point intensity value of the corrected image S.sub.corr(t) for sensitive element i,j is then

[00004]$S_{i,j_{\mathrm{corr}}}\left(t\right)=\frac{a}{1-\exp \left(-\frac{\Delta }{\tau }\right)}.Math.\left[S_{i,j\_\mathrm{raw}}\left(t\right)-e^{-\Delta /\tau }.Math.S_{i,j\_\mathrm{raw}}\left(t-\Delta \right)\right].$

The use of such value for coefficient β makes it possible to reduce attenuation effects on the intensity of the corrected image when the duration A between the respective read-out times of the two raw images is short.

[0047] The expressions cited above for coefficient a, and possibly also for coefficient β, make it possible to correct the memory effect in a particularly efficient manner when the sensitive element is of a type with first-order transfer function with respect to time, as described above. Indeed, in this case, the raw image S.sub.raw(t-Δ), corresponding to the prior raw image as called in the general part of the present description, when it is multiplied by α=exp(-Δ/τ), quantifies the memory effect contribution which is associated with all the radiation received by each sensitive element before the capture of this prior raw image. This memory effect contribution, which can be called the long-term memory effect and which contributes to the new raw image S.sub.raw(t), is then completely eliminated by the invention in the corrected image S.sub.corr(t). However, another memory effect contribution, which is associated with the radiation received by each sensitive element during the duration Δ between the respective read-out times of both raw images, remains. This can be called the short-term memory effect.

[0048] When a sensitive element is not of a type with first-order transfer function with respect to time, the expression of coefficient α as a function of the characteristic response time τ and of the duration Δ can still be used. To this end, an empirical value adapted to the sensitive element concerned may be adopted for the characteristic response time τ, even if this value has no theoretical significance relating to the transfer function f(s) of the sensitive element.

[0049] [FIG. 1 2] illustrates a video sequence capture, during which raw images are periodically captured at read-out times separated by a duration Δ. Thus, a last raw image S.sub.raw(t) is composed of the raw detection signals S.sub.i,j_raw(t) read at time t, an immediately prior raw image S.sub.raw(t-Δ) is composed of the raw detection signals S.sub.i,j_raw(t-Δ) read at time t-Δ, a raw image S.sub.raw(t-2Δ) preceding that one is composed of the raw detection signals Si,.sub.i raw(t-2Δ) read at time t-2Δ, yet another raw image S.sub.raw(t-3Δ) preceding that one consists of the raw detection signals S.sub.i,j_raw(t-3Δ) read at time t-3Δ, etc. Then a corrected video sequence, composed of images corrected according to the invention, can be constructed with a same frame rate equal to 1/Δ, by combining each raw image with the last one read just before it. Thus, raw images S.sub.raw(t) and S.sub.raw(t-Δ) are combined in order to obtain according to the invention the corrected image S.sub.corr(t), composed of the image point intensity values S.sub.i,j_corr(t); raw images S.sub.raw(t-Δ) and S.sub.raw(t-2Δ) are combined in order to obtain according to the invention the corrected image S.sub.corr(t-Δ), composed of the image point intensity values S.sub.i,j_corr(t-Δ); raw images S.sub.raw(t-2Δ) and S.sub.raw(t-3Δ) are combined in order to obtain according to the invention corrected image S.sub.corr(t-2Δ), composed of the image point intensity values S.sub.i,j_corr(t-2Δ), etc. The axis on the left in [FIG. 2] corresponds to the time coordinate, denoted t. The image sensor of the invention may then comprise a display system which is controlled so as to display the corrected images, and possibly also the raw images according to a temporal correspondence. An improvement in the image contrast and tail attenuation which result from correction of the long-term memory effect can thus be assessed. This mode of correcting images in a video sequence is particularly suitable for obtaining a sequence of corrected images at the maximum image capture rate possible for the matrix 10a and the controller 10b.

[0050] [FIG. 3] corresponds to [FIG. 2] when each corrected image is obtained by combining two successive raw images with the pairs of raw images dedicated to different corrected images, that are disjoint, non-interlaced, and successive. Thus, raw images S.sub.raw(t) and S.sub.raw(t-Δ.sub.1), which are composed of raw detection signals S.sub.i,j_raw(t) and S.sub.i,j_raw(t-Δ.sub.1)), are combined in order to obtain according to the invention corrected image S.sub.corr(t). Similarly, raw images S.sub.raw(t-(Δ.sub.1+Δ.sub.2)) and S.sub.raw(t-(2Δ.sub.1+Δ.sub.2)), which are composed of raw detection signals S.sub.i,j_raw(t-(Δ.sub.1+Δ.sub.2)) and S.sub.i,j_raw(t-(2Δ.sub.1+Δ.sub.2)), are combined in order to obtain according to the invention corrected image S.sub.corr(t-(Δ.sub.1+Δ.sub.2)), composed of image point intensity values S.sub.i,j_corr(t-(Δ.sub.1+Δ.sub.2)), etc. Ai is the duration between the read-out times for both raw images of a same pair, for obtaining the corresponding corrected image, and Δ.sub.2 is the duration between the read-out times of the prior raw image of the last pair and of the new raw image of the previous pair. The video frequency of the corrected images is then 1/(Δ.sub.1+Δ.sub.2). This other mode of correcting images in a video sequence is suitable for obtaining a video of corrected images at low frame rate, while being able to use a short duration between the two raw images combined to obtain each corrected image.

[0051] [FIG. 4] also corresponds to [FIG. 2], again considering a video sequence of raw images which are captured at frequency 1/Δ. But this time, a series of corrected images is produced by each time combining the same last captured raw image with a different raw image by chronologically and progressively using the prior raw images upstream in the video sequence. Thus, a first corrected image S.sub.corr_1 composed of image point intensity values S.sub.i,j_corr_1, is obtained by combining raw images S.sub.raw(t) and S.sub.raw(t-Δ) separately constructed from raw detection signals S.sub.i,j_raw(t) and S.sub.i,j_raw(t-Δ), using coefficients

[00005]$\alpha =\exp \left(-\frac{\Delta }{\tau }\right)\mathrm{and}\beta =\frac{a}{1-\exp \left(-\frac{\Delta }{\tau }\right)}.$

A second corrected image S.sub.corr_2, composed of image point intensity values S.sub.i,j_corr_2, is obtained by reusing raw image S.sub.raw(t) and combining it with image S.sub.raw(t-2Δ) composed of raw detection signals S.sub.i,j_raw(t-2Δ), using coefficients

[00006]$\alpha =\exp \left(-\frac{2.Math.\Delta }{\tau }\right)\mathrm{and}\beta =\frac{a}{1-\exp \left(-\frac{2.Math.\Delta }{\tau }\right)}.$

A third corrected image S.sub.corr_3, composed of image point intensity values S.sub.i,j_corr_3, is obtained by combining raw image S.sub.raw(t) with image S.sub.raw(t-3Δ) constructed from raw detection signals S.sub.i,j_raw(t-3Δ), using coefficients

[00007]$\alpha =\exp \left(-\frac{3.Math.\Delta }{\tau }\right)\mathrm{and}\beta =\frac{a}{1-\exp \left(\frac{3.Math.\Delta }{\tau }\right)}.$

Similarly, a fourth corrected image S.sub.corr_4, composed of image point intensity values S.sub.i,j_corr_4, is obtained by combining raw image S.sub.raw(t) with image S.sub.raw(t-4Δ) constructed from raw detection signals S.sub.i,j_raw(t-4Δ), using coefficients

[00008]$\alpha =\exp \left(-\frac{4.Math.\Delta }{\tau }\right)\mathrm{and}\beta =\frac{a}{1-\exp \left(-\frac{4.Math.\Delta }{\tau }\right)},$

etc. The set of corrected images thus obtained makes it possible to assess the effects of the duration between the raw images which are combined according to the invention, on the quality of the corrected images. Some of these advantageous effects are presented here: better rendering of high spatial frequencies in the image in the presence of lateral movement, and reduction of the tail effect.

[0052] Rendering of image spatial frequencies in the presence of lateral movement.

[0053] It is known that one way to highlight the attenuation of the temporal transfer function of a matrix of sensitive elements is to capture an image of a spatial pattern which is periodic and moved at constant speed parallel to its direction of periodicity. Each sensitive element thus receives radiation whose instantaneous intensity varies periodically according to a temporal frequency value which is equal to the product of the apparent travel speed and the period of the pattern. A scene which consists of bands parallel to the direction of the columns of the matrix 10a, and whose luminance varies sinusoidally parallel to the direction of the rows of the matrix 10a, is imaged onto the matrix 10a of sensitive elements 1. The spatial frequency of these bands in the scene image is denoted v.sub.s, which can be expressed in pixels.sup.−1. This scene is moving at constant speed parallel to the rows of the matrix 10a, and V designates the speed of movement of the image of the scene on the matrix 10a, which can be expressed in pixels/s (pixels per second). It is assumed that all the sensitive elements 1 have the same value for the characteristic response time τ, and that their common transter tunction is

[00009]$f\left(s\right)=\frac{G}{1+\tau .Math.s}.$

[FIG. 5] is a diagram which illustrates the variations of the apparent transfer function of the image sensor, as a function of the spatial image frequency v.sub.s or of the movement speed V. The horizontal axis of the diagram identifies the values of the spatial frequency v.sub.s, multiplied by the characteristic response time τ and by the speed V, in order to obtain a dimensionless variable. The product V.Math.vs corresponds to the temporal frequency of variation of the intensity of the radiation received by each sensitive element. The vertical axis identifies the values F of the amplitude of the apparent transfer function of the image sensor. The lower curve which appears in [FIG. 5], and which is designated by the words “without correction”, corresponds to the raw detection signals S.sub.i,j_raw as delivered by the controller 10b, before memory effect correction, as functions of the speed V or of the spatial frequency v.sub.s. It is therefore the read-out signals which compose each raw image captured during movement of the scene. The curve denoted Δ=τ corresponds to the image point intensity values of a corrected image which is obtained by combining two raw images captured during movement of the scene at read-out times separated by a duration Δ which is equal to τ. Similarly, the curve which is denoted Δ=τ/4 corresponds to the image point intensity values of a corrected image which is obtained by combining two raw images captured during movement of the scene at read-out times separated by a duration A which is equal to τ/4. These corrected images use values for coefficient β which are calculated for each one as indicated above in this description. The general increase in amplitude F of the transfer function of the image sensor when going from the raw image to the corrected image with Δ=τ, then to the corrected image with Δ=τ/4, corresponds to an increasing correction of the memory effect of the sensitive elements 1. In other words, the short-term memory effect which is not compensated for in the corrected images is increasingly reduced. The gain in amplitude F of the transfer function of the image sensor, compared to the raw image when the dimensionless spatial frequency is greater than 1.0, is about 2.1 for the image corrected with Δ=τ, apart from particular spatial frequencies corresponding to integer values for v.sub.s.Math.τ.Math.V. For these particular spatial frequencies, called blind frequencies, both raw images which are combined to obtain the corrected image are identical, so that the memory effect correction has no effect. For the image corrected with Δ=τ/4, the gain in amplitude F of the transfer function of the image sensor, again in comparison to the raw image but when the dimensionless spatial frequency is greater than 2.0, is about 8, and the blind frequencies correspond to values for v.sub.s.Math.τ.Math.V which are multiples of four.

[0054] Tail effect.

[0055] The matrix 10a used comprises 320 columns and 240 rows of sensitive elements 1, and the characteristic response time τ of all the sensitive elements is approximately 14 ms. An image sensor which is composed from this matrix of sensitive elements captures a video sequence having a uniform background composed of a black body at 325 K (Kelvin), through the upper part of an opaque rotating disc which has three radial slit openings. The matrix 10a is optically conjugate with the rotating disk, and the axis of rotation of the disk is parallel to the optical axis of the conjugating optics used. The rotation speed of the disk is 1.5 revolutions per second, and the raw image acquisition rate is 60 images per second, corresponding to a duration between the respective read-out times for any two successive raw images which is equal to 16.7 ms. The diagram of [FIG. 6] reproduces the values of the raw detection signals which are read in a row of the matrix 10a for two raw images successively captured in the video, respectively denoted S.sub.raw(t-Δ) and S.sub.raw(t). It also reproduces the image point intensity values for the same row of the matrix 10a, for the corrected image S.sub.corr(t) which is obtained according to the invention from the two raw images S.sub.raw(t-Δ) and S.sub.raw(t). The movement of one of the radial slits of the rotating disc is visible between both raw images: from the right of the diagram to the left. The horizontal axis of the diagram of [FIG. 6] identifies, by their column numbers nc, the sensitive elements 1 of the row under consideration in the matrix 10a, and the vertical axis identifies the values I of the raw detection signals or the image point intensity values, for the three images S.sub.raw(t), S.sub.raw(t-Δ) and S.sub.corr(t). The progressively inclined rising and falling edges of the intensity curve, which correspond to the two edges of the slit in the raw images S.sub.raw(t) and S.sub.raw(t-Δ), constitute a tail effect due to the combination of movement of the slit and the memory effect of the sensitive elements 1. In the corrected image S.sub.corr(t), the two edges of the slit appear sharper, showing the efficiency of the memory effect compensation. For this example, the formulae provided above in the present description for coefficients α and β result in S.sub.i,j_corr(t)=1.437×S.sub.i,j_raw(t)-0.437×S.sub.i,j_raw(t-Δ). Generally, the adjustment of the ratio between the duration Δ which separates successive raw images of a video sequence and the characteristic response time τ of the sensitive elements may depend on a trade-off between reduction of the tail effect, improvement in the geometric rendering of the scene elements in the images, and amplification of image noise in the corrected images. For example, values that are between 0.02 and 0.2 for the ratio Δ/τ may be adapted when the signal-to-noise ratio of the raw image is greater than 150, respectively 15.

[0056] It is understood that the invention can be reproduced while modifying secondary aspects of the modes of implementation which have been described in detail above, while retaining at least some of the advantages cited. In particular, the selection of the raw images which are combined in pairs to obtain the corrected images, can be modified with respect to the illustrated examples. In addition, one will recall that the use of a coefficient β which depends on the duration Δ between the raw images combined to form a corrected image is optional, even if the coefficient a used is calculated according to this duration Δ.