Method for correcting defects in an image provided by an image sensor and in particular for reducing noise

Abstract

The invention relates to a method for correcting defects occurring in an image generated by an image sensor (IS), the method comprising the following steps: receiving an image to be corrected (IM) captured by the image sensor; receiving from the image sensor a temperature (TP) obtained when the image sensor captured the image to be corrected; receiving an integration time (IT) applied by the image sensor when capturing the image to be corrected; and, for each pixel (px[i,j]) in the image to be corrected, subtracting from the pixel value a pixel-specific noise correction coefficient (bm[i,j]) derived from a noise reduction model, the noise reduction model comprising a linear component depending on the temperature of the image sensor plus an exponential component depending on the temperature of the image sensor multiplied by the integration time, the linear and exponential components depending on pixel-specific coefficients (ab[i,j], bb[i,j], ad[i,j], bd[i,j]).

Description

Technical Field

[0001] The present invention relates to the field of imaging devices, such as video and still cameras. In particular, the invention relates to imaging devices in which an image sensor is integrated into CMOS (Complementary Metal Oxide Semiconductor) or CCD (Charge Coupled Device) technology, either of the CTIA (Charge Transimpedance Amplifier) ​​or SF (Source Follower) type. The invention can be applied to imaging in the visible range as well as in the SWIR, MWIR, and LWIR (Short Wave, Medium Wave, and Long Wave Infrared) ranges. Background Art

[0002] Typically, a CMOS image sensor consists of pixels or photosites arranged in an array configuration. Each pixel includes a photosensitive area, typically a photodiode, which is configured to accumulate charge based on the light it receives; and a readout circuit for measuring the amount of charge accumulated by the photodiode. The readout circuit includes a transfer transistor to control the transfer of the charge accumulated in the photodiode to a readout node. Therefore, the pixels are controlled according to a cycle that includes an initialization phase, an integration phase, and a readout phase. During the integration phase, the photodiode accumulates charge based on the light it receives. The readout phase includes generating a signal corresponding to the amount of charge accumulated by the photodiode during the integration phase. The initialization phase includes eliminating the charge accumulated by the photodiode during the integration phase.

[0003] The images generated by today's image sensors are subject to interference from various noise sources, particularly as image sensors become smaller. The effects of some of these noise sources appear in captured images when the image sensor is placed in darkness. Under these conditions, the image sensor's pixel circuitry generates signals despite the pixel being completely dark. It turns out that the amplitude of these dark signals varies with the image sensor's temperature and the integration time chosen to generate the image. The amplitude of these dark signals can also vary from one pixel circuit to another within the same image sensor, as well as from one image sensor to another, even if they come from the same manufacturing batch.

[0004] It is known to create a fixed-pattern noise image by generating a dark image using an image sensor placed in darkness. This dark image is then subtracted from the image generated by the image sensor. To account for temperature variations and integration time, this dark image must be generated each time a new image is acquired. However, in many applications, particularly video capture, the time available before each frame is often insufficient to generate this dark image. Furthermore, in the case of video images, correcting for each frame or frame sequence inevitably introduces potentially annoying flicker.

[0005] Therefore, it is desirable to provide an effective noise reduction method in an imaging device that takes into account changes in image sensor temperature and integration time and does not hinder the viewing experience. It is also desirable that the noise reduction method be applicable to each imaging device. Summary of the Invention

[0006] An embodiment relates to a method for correcting defects appearing in an image generated by an image sensor, the method comprising the steps of: receiving an image to be corrected captured by the image sensor; receiving a temperature from the image sensor obtained when the image to be corrected was captured; receiving an integration time applied by the image sensor when capturing the image to be corrected, and subtracting, for each pixel in the image to be corrected, from the pixel value a pixel-specific noise correction factor obtained from a noise reduction model, the noise reduction model comprising a linear component depending on the temperature of the image sensor plus an exponential component depending on the temperature of the image sensor multiplied by the integration time, the linear component and the exponential component depending on pixel-specific coefficients.

[0007] According to one embodiment, the noise reduction model is defined by the following equation:

[0008] bm[i,j]=IT×ad[i,j]×Exp(bd[i,j]×TP)+ab[i,j]×TP+bb[i,j]

[0009] where bm[i,j] is the noise correction factor to be subtracted from the corresponding pixel of the image to be corrected, IT is the integration time, TP is the temperature of the image sensor, EXP is the exponential function, and ad[i,j], bd[i,j], ab[i,j], and bb[i,j] are pixel-specific coefficients.

[0010] According to one embodiment, the method comprises calculating a noise correction factor for each pixel each time the integration time is changed or each time the temperature of the image sensor deviates from a previous value by more than a temperature deviation threshold.

[0011] According to one embodiment, the method includes the following steps: obtaining images by an image sensor with a minimum integration time in the absence of light, each image being taken at a different corresponding temperature; and determining, for each pixel of the image to be corrected, a coefficient of a linear component by a linear regression calculation applied to the corresponding pixel in the image taken at the corresponding temperature in the absence of light.

[0012] According to one embodiment, the method includes the following steps: obtaining images by an image sensor with different integration times in the absence of light, the image sensor being subjected to different temperatures; generating a correction image obtained by subtracting an image captured by the image sensor in the absence of light at the same temperature with a minimum integration time from each image captured by the image sensor in the absence of light with different integration times at different temperatures; and determining a coefficient of an exponential component by an exponential fitting calculation applied to corresponding pixels in the correction images obtained at different temperatures and corresponding to the same integration time.

[0013] According to one embodiment, the coefficients of the exponential component are determined by averaging the coefficients obtained by exponential fitting calculations for different integration times.

[0014] According to an embodiment, the noise reduction model comprises the same components for all pixels of the image sensor that depend on the integration time.

[0015] According to one embodiment, the method includes obtaining a video image stream, wherein a noise correction factor corresponding to each pixel in an image of the video stream is subtracted from the corresponding pixel in each image of the video stream.

[0016] According to one embodiment, the method includes the following steps: receiving a command to select a gain value for an image sensor; and selecting a pixel-specific noise correction factor for each pixel in an image to be corrected based on the selected gain value, the noise correction factor being used to correct the value of each pixel in the image to be corrected, the noise correction factor being determined based on a set of pixel-specific coefficients generated according to the selected gain value.

[0017] According to one embodiment, the method includes: for each pixel in the corrected image after noise reduction, multiplying the value of the pixel by a pixel-specific gain correction factor obtained from a gain normalization table to obtain an image with normalized gain.

[0018] According to one embodiment, the method includes updating a gain normalization table each time an integration time is changed or each time a temperature of the image sensor deviates from a previous value by more than a temperature deviation threshold, the updating of the gain normalization table being implemented by an interpolation calculation applied to a set of gain normalization tables determined for different temperatures.

[0019] According to one embodiment, the method includes the following steps: obtaining, by an image sensor, a sequence of images with different integration times or at different intensities of the light source in the presence of a uniform light source, each image sequence being taken at a different corresponding temperature; determining, for each pixel in an image in each image sequence, a gain by a linear regression calculation applied to the corresponding pixel in the image in the image sequence; and determining, for each pixel in an image in each image sequence, a gain correction factor by dividing the average of the gains obtained for all pixels of the image sequence by the gain determined for the pixel.

[0020] According to one embodiment, the method comprises the following steps: obtaining a sequence of images by an image sensor in the absence of light with an average integration time, wherein each image is obtained with an image sensor subjected to a corresponding temperature; calculating, for each pixel in an image in the obtained sequence of images, an average value of deviations between values ​​of the pixels of the images in the sequence of images corresponding to the image sensor temperature at different image sensor temperatures and a noise correction factor defined for the pixel at the image sensor temperature with the average integration time, each deviation being calculated for one image sensor temperature; and comparing the average value of the deviations with a threshold value, and if the average value of the deviations is greater than the threshold value for the pixel, the pixel is deemed defective.

[0021] According to an embodiment, the method includes the following steps: correcting each image obtained by the image sensor by replacing the value of the defective pixel with the value of the adjacent pixel or the average value of the adjacent pixels, or correcting each image obtained by the pixel sensor by replacing the corresponding values ​​of the defective pixel and the pixels adjacent to the defective pixel with the values ​​of the pixels adjacent to the defective pixel and the pixels adjacent to the defective pixel or the average value of the pixels adjacent to the defective pixel and the pixels adjacent to the defective pixel.

[0022] Embodiments may also relate to a device for correcting defects occurring in an image generated by an image sensor, the device being configured to implement the method defined previously.

[0023] Embodiments may also relate to an imaging device comprising an image sensor, circuitry for obtaining a temperature of the image sensor, and circuitry for obtaining an integration time applied to the image sensor, the imaging device being configured to implement the previously defined method.

[0024] According to one embodiment, the image sensor is of CTIA or SF type.

[0025] The embodiments may also relate to a computer program product loadable into the memory of a computer, which, when executed by the computer, configures the computer to perform the method previously defined. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Exemplary embodiments of the present invention will become more apparent from the following description which is provided for illustrative purposes only and is shown in the accompanying drawings.

[0027] FIG1 schematically shows a conventional imaging device.

[0028] Figure 2 schematically illustrates an image sensor associated with a noise reduction device for an imaging device according to one embodiment,

[0029] Figure 3 Schematically illustrates a method for calculating a noise reduction coefficient according to an embodiment,

[0030] Figures 4 to 6 The curve showing the change of pixel signal intensity with the temperature of the image sensor,

[0031] Figure 7 Schematically shows a block diagram of a gain correction circuit for receiving an image from a noise reduction device according to one embodiment,

[0032] Figure 8 The figure schematically shows a block diagram of a circuit for generating a gain correction coefficient table according to one embodiment. DETAILED DESCRIPTION

[0033] FIG1 shows an image sensor IS1 having processing circuits ADC, AMP, and PRC. The image sensor IS1 can be integrated into a portable device, such as a camera, camcorder, mobile phone, or any other device with image capture functionality. The image sensor IS1 typically includes an array PXA of pixel circuits PC. Array PXA includes pixel circuits PC arranged in multiple rows and columns. The image sensor IS1 also includes control circuits RDRV, RDEC, CDRV, CDEC, TAC, which are configured to provide different control signals to the pixel circuits PC depending on the different stages to be linked to capture an image. The readout and processing circuitry may include an amplifier AMP, an analog-to-digital converter ADC, and a processor PRC. Array PXA provides pixel signals to the readout and processing circuits AMP, ADC, and PRC, which are configured to provide an image IM based on the pixel signals.

[0034] Pixel rows are selectively activated by row driver circuit RDRV in response to row address decoder RDEC. Pixel columns, in turn, are activated by column driver circuits controlled by column decoder CDEC. Circuits RDRV and CDRV provide appropriate voltages to drive pixel circuits PC. Sensor IS also includes control circuit TAC, which drives address decoders RDEC and CDEC and driver circuits RDRV and CDRV to select the appropriate pixel row and column for pixel readout at any given time. The read pixel signal is amplified by amplifier AMP and converted to a digital signal by analog-to-digital converter ADC. The digitally converted pixel signal is processed by image processor PRC, which provides an image IM based on the digitally converted pixel signal. Image processor PRC may include memory for processing and storing the received image signal.

[0035] Figure 2 FIG. 1 shows an image sensor IS associated with a noise reduction device according to an embodiment. Figure 2 In an example, a noise reduction device includes a noise reduction function implemented by a processor IPRC coupled to a memory MEM. The processor IPRC may be integrated with or connected to an image sensor IS. The noise reduction function is defined based on a noise model having two components: a bias component that varies linearly with the temperature of the image sensor, and a dark current component that varies with temperature and integration time.

[0036] According to one embodiment, the noise model implemented by the processor IPRC is defined by the following equation:

[0037] bm[i,j]=IT×ad[i,j]×Exp(bd[i,j]×TP)+ab[i,j]×TP+bb[i,j] (1)

[0038] Where bm[i,j] is the pixel of the correction image BM to be subtracted from each image generated by the image sensor IS, the position of the pixel is identified by the row and column indices i, j, IT is the integration time, TP is the temperature measured by the image sensor IS or a temperature sensor TS coupled to the image sensor, ad[i,j], bd[i,j], ab[i,j], and bb[i,j] are coefficients determined for the pixel to be corrected at position [i,j], and EXP is an exponential function. The coefficients ad[i,j], bd[i,j], ab[i,j], and bb[i,j] are input and stored as tables AD, BD, AB, BB in the memory MEM of the imaging device. Each pixel px[i,j] of the image IM generated by the image sensor is corrected to generate a corrected image OIM by applying the following formula:

[0039] px'[i,j]=px(j,j)-bm[i,j] (2)

[0040] where px'[i,j] is the corrected value of pixel px[i,j].

[0041] Under these conditions, the pixel values ​​provided by the image sensor IS can be individually corrected for all integration times and operating and imaging device temperatures. Due to the simplicity of the correction by subtracting from each pixel px[i,j] of the image IM the corresponding pixel bm[i,j] of the correction image BM, the images provided by the image sensor IS can be processed at very high frame rates. Thus, the processor IPRC can process video streams, including those of several hundred frames per second. If necessary, this correction can be performed by hard-wired logic circuits (such as Figure 2 In addition, by providing such a circuit for each pixel of a frame, all pixels of each frame can be processed in parallel at the same time.

[0042] Similarly, due to the simplicity of equation (1) involving only addition, multiplication and exponential EXP operations, the calculation of the correction image BM can be performed in real time after a change in temperature TP or a change in integration time IT. If necessary, the update function of the correction image BM performed by the processor IPRC can also be replaced by a hard-wired logic circuit (such as Figure 2 As shown in Figure 2, the exponential operation EXP can be performed simply by a lookup table. Since there is no interaction between the pixels of the known image BM, all pixels bm[i,j] of the corrected image BM can also be calculated in parallel.

[0043] Figure 3 A circuit TGC for generating coefficient tables AD, BD, AB, and BB is shown. During a calibration phase, circuit TGC subjects image sensor IS to different temperatures TP1, TP2, ..., TPm and, for each temperature, sequentially sets integration time IT to different values ​​IT0, IT1, ..., ITn, including the minimum value IT0. Circuit TGC receives as input a dark image DIM [IT0-ITn, TP1-TPm] obtained using an imaging device. Temperatures TP1-TPm can be selected within the operating temperature range of image sensor IS. Furthermore, the number of integration time values ​​can be set to a number between 5 and 15 within the range of possible integration time values ​​for the imaging device.

[0044] In a first step, the bias coefficients ab[i,j], bb[i,j] are calculated from the image DIM[IT0,TP1-TPm] obtained with the minimum integration time IT0, where in equation (1), the integration time IT is a multiplication factor in the dark signal component, so this component can be ignored when the integration time is very small. According to one example, the minimum integration time IT0 is between 10 and 100 μs, and preferably between 40 and 60 μs. As a first approximation, each pixel dx[i,j,IT0,TP] of the image DIM[IT0,TP1-TPm] obtained with the integration time IT0 can be modeled by the following formula:

[0045]

[0046] Where TP = TP1, TP2, ... TPm. Therefore, by calculating LR through linear regression, the bias coefficients ab[i,j] and bb[i,j] in Tables AB and BB can be determined for each pixel dx[i,j,IT0,TP] in the temperature range TP1-TPm.

[0047] In a second step, the darkness coefficients ad[i,j], bd[i,j] of the tables AD, BD are calculated. To this end, the corrected image DIM'[IT1-ITn,TP1-TPm] is obtained from the image DIM[IT1-ITn,TP1-TPm] by subtracting from each pixel dx[i,j,IT,TP] in each image DIM[IT,TP] obtained for the integration times IT1-ITn and temperatures TP1-TPm the value of the corresponding pixel dx[i,j,IT0,TP] in the image DIM[IT0,TP] obtained for the same temperature TP and for the minimum integration time IT0. Thus, considering equation (1), each pixel dx'[i,j,IT,TP] in the corrected image DIM[IT1-ITn,TP1-TPm] can be modeled by the following equation:

[0048] dx'[i,j,IT,TP]=IT×ad[i,j,IT]×Exp(bd[i,j,IT]×TP) (4)

[0049] Among them, dx'[i,j,IT,TP]=dx[i,j,IT,TP]-dx[i,j,IT0,TP].

[0050] The dark coefficients ad[i,j] and bd[i,j] can be determined for each integration time IT1-ITn by calculating EF by exponential fitting. This calculation is done, for example, The dark coefficients ad[i,j] and bd[i,j] stored in tables AD and BD can be obtained by averaging the coefficients ad[i,j,IT] and bd[i,j,IT] obtained with integration times IT1-ITn, respectively.

[0051] Depending on the image sensor, it may be appropriate to add the same compensation component to the noise model for all pixels of the image sensor, depending on the integration time. This component can be determined by comparing the pixel at position [i, j] in the image DIM obtained at the same temperature TP for the integration times IT0-ITn. This comparison can be performed by considering several pixels in each image DIM obtained at the same temperature TP for different integration times IT0-ITn, wherein the compensation component is set to the average value of the values ​​obtained for the considered pixels.

[0052] Figures 4 to 6 Each of them represents a curve C2, C4, C6 of the variation of the value of the pixel px[i,j] as a function of the temperature TP in the uncompensated image IM provided by the image sensor IS, and a curve C1, C3, C5 corresponding to the model bm[i,j] defined by equation (1) and determined for the pixel [i,j] by the coefficients ad[i,j], bd[i,j], ab[i,j] and bb[i,j]. The curves C2, C4, C6 are determined according to the temperature TP. and The pixel value px[i,j] obtained at is drawn.

[0053] Figure 4 The curves C1 and C2 in FIG. 5 correspond to an integration time of 50 μs. Figure 4 A difference of less than 0.6% is shown between the value of a pixel px[i,j] at the output of the image sensor IS and the corresponding value bm[i,j] determined by the model (Equation (1)).

[0054] Figure 5 and 6 The C3-C6 curves in FIG were obtained with integration times of 6.654 ms and 13.321 ms, respectively. Figure 5 A maximum deviation of less than 14% between the measured value of a pixel px[i,j] and the corresponding value bm[i,j] determined by the model is shown. Figure 6 The maximum deviations of less than 8% between the values ​​of pixels px[i,j] output by the image sensor IS and the corresponding values ​​bm[i,j] determined by the model are shown. It can be seen that these maximum deviations are obtained at temperatures above 38°C.

[0055] According to an embodiment, the processor IPRC periodically or continuously receives the temperature TP from the image sensor IS. Each time the processor IPRC calculates the correction image BM, it stores the current temperature measured by the image sensor IS. When the current temperature measurement provided by the image sensor deviates from the stored value, the processor IPRC recalculates the correction image BM, taking into account the last temperature measurement provided by the image sensor. Depending on the implementation, a new calculation of the correction image BM occurs when a temperature difference between 0.5°C and 2°C (e.g., 1°C) occurs. Similarly, when the IT integration time is changed, the IPRC processor recalculates the correction image BM based on the new integration time.

[0056] According to one embodiment, the image sensor IS has several gain values ​​that can be selected from the control interface of the imaging device or based on the illumination conditions of the image sensor. In this case, a dark image DIM is generated for each gain value to determine a table of coefficients AD, BD, AB, and BB for each gain value. The integration time IT used to generate the image DIM can be selected based on the gain value to avoid the unlikely scenario of a high risk of saturation of the image sensor IS. Furthermore, the processor IPRC calculates a current corrected image BM for each gain value based on the integration time IT and the temperature of the image sensor IS.

[0057] Due to variations in gain from one pixel to another and depending on the temperature of the image sensor, the quality of the image generated by the image sensor may also degrade. Gain variations from one pixel to another may be caused in particular by structural differences between the pixel circuits.

[0058] Figure 7 A gain correction circuit according to an embodiment is shown. The gain correction circuit receives the corrected image OIM at the output of the noise reduction circuit and multiplies each pixel px'[i,j] of the corrected image OIM by a gain correction factor gf[i,j] calculated for pixel [i,i] based on the current temperature of the image sensor. As a result, the pixels px"[i,j] of the resulting corrected image GCI are generated with a uniform gain across the entire image GCI.

[0059] The gain correction factor gf[i,j] belongs to a gain normalization table GF[i,j], which is stored in a memory MEM and determined by an interpolation module ITP based on the image sensor temperature TP and a set of gain normalization tables GF[TP1], ...GF[TPm] determined for various temperatures TP1, ...TPm. The interpolation applied by the interpolation module ITP can be, for example, linear or polynomial interpolation.

[0060] According to an embodiment, the gain normalization table GF is updated when a temperature difference occurs between 0.5° C. and 2° C. (for example 1° C.) relative to the temperature previously considered for the calculation of the correction table.

[0061] Figure 7 The image processing shown in can be implemented by the processor IPRC to process video streams, including video streams of several hundred frames per second. If necessary, this processing can be performed by hard-wired logic circuits (such as Figure 7 In addition, by providing such a circuit for each pixel in a frame, all pixels in each frame can be processed in parallel at the same time.

[0062] Similarly, due to the simplicity of the processing involving only addition, multiplication and inversion operations, the calculation of the gain normalization table GF can be performed in real time after the temperature change. If necessary, the function performed by the processor IPRC for updating the gain normalization table GF can also be replaced by a hard-wired logic circuit (such as Figure 7 As shown), it is known that the interpolation operation can be performed at least partially using one or more lookup tables if necessary. All pixels gf[i,j] of the gain normalization table GF can also be calculated in parallel.

[0063] Figure 8 A circuit GGC is shown for generating a gain normalization table GF[TP1], ...GF[TPm] according to an embodiment. The calculation circuit receives as input an image UIM[IT0 ... ITn, TP1 ... TPm] obtained during a calibration phase, using an imaging device placed in front of a light source having a substantially uniform light intensity in all directions. The images UIM[IT0 ... ITn, TP1 ... TPm] are obtained by subjecting the image sensor IS to different temperatures TP1, ...TPm and, for each temperature, by successively setting the integration time IT to a different value IT0, ...ITn. The light source used to generate the images UIM[IT0 ... ITn, TP1 ... TPm] may be an integrating sphere, with the imaging device placed within the aperture of the sphere. Instead of varying the integration time IT, the light intensity emitted by the uniform light source can also be varied by setting the integration time of the image sensor to, for example, an average value. The temperatures TP1, ...TPm used are set, for example, to 20, 30, 40, and 50°C.

[0064] Circuit GGC calculates a gain table PG[TP] for each of the temperatures TP1-TPm using a linear regression calculation circuit RL, where the gain pg[i,j,TP] for each pixel [i,j] corresponds to the average slope of the curve of the value of the pixel px[i,j] as a function of the integration time IT. By calculating the average AV, circuit GGC then determines an average gain table PGM[TP] for each temperature TP=TP1, ...TPm, including the average gain value for each pixel [i,j]. Each gain normalization table GF[TP1], ...GF[TPm] is then generated by dividing the corresponding value pgm[i,j] in the average gain table PGM[TP] by the corresponding value pg[i,j,TP] in the gain table PG[TP] for each pixel [i,j].

[0065] The quality of the image provided by the image sensor IS may also be affected by the presence of defective pixel circuits. According to one embodiment, the processor IPRC is configured to calculate for each pixel [i, j] the average integration time IT at different temperatures TP = TP1, ... TPm. moy The average value of the deviation E[i,j] between the pixel value dx[i,j,IT,TP] of the image DIM provided by the image sensor IS and the correction value bm[i,j,IT,TP] for that pixel is calculated and compared with a threshold value. If this average deviation is greater than the threshold value for pixel [i,j], pixel [i,j] is considered defective. Detection of defective pixels can be performed during a calibration phase after obtaining the image DIM. For example, the average deviation E[i,j] can be calculated for each pixel [i,j] using the following formula:

[0066] E[i,j]=[∑ IP (dx[i,j,IT moy ,TP]-bm[i,j,IT moy ,TP]) 2 ] 1 / 2 (5)

[0067] Among them, TP=TP1, TP2...TPm.

[0068] According to one embodiment, the processor IPRC performs correction of the image OIM or GCI by replacing the value px'[i,j] or p"[i,j] of each pixel thus detected as defective with the value of the adjacent pixels or the average value of the pixels adjacent to the defective pixel. Pixels adjacent to the defective pixel may also be considered defective and form a defective pixel area, in which case each pixel in the defective pixel area may be replaced with the adjacent pixels of the defective pixel area or the average value of these adjacent pixels.

[0069] It will be clear to those skilled in the art that the present invention is susceptible to variations and various applications. In particular, the present invention is not limited to image sensors that perform the calculation of the correction image BM. In practice, the correction image BM can be determined by an external computer having tables AB, BB, AD, BD and receiving the integration time IT and the temperature TP of the image sensor IS from the image sensor, and, if necessary, by the gain applied by the image sensor.

[0070] Furthermore, the correction of the image provided by the image sensor may also be performed by such an external computer which receives the uncompensated image IM from the image sensor.

[0071] Additionally, the temperature TP is not necessarily provided by the image sensor IS, but may be measured by an external temperature sensor associated with the image sensor.

[0072] The processed image can be a still image or a video image.

[0073] Pixel gain correction can be performed from the corrected image OIM obtained by other noise reduction processes, so that the pixel gain correction can be performed without performing a reference Figure 2 Gain correction is implemented in the context of the described noise reduction correction.

[0074] Similarly, the detection of defective pixels as described above can be performed without using Figure 2 and Figure 7 The correction of defective pixels can be performed on an uncorrected image, i.e. without using any of the correction methods shown in FIG. Figure 2 and 7 Any of the correction methods shown.

Claims

1. A method for correcting defects occurring in an image generated by an image sensor, the method comprising the steps of: receiving an image to be corrected captured by the image sensor, receiving, from the image sensor, a temperature obtained when the image sensor captures the image to be corrected, receiving an integration time applied by the image sensor when capturing the image to be corrected, subtracting, for each pixel in the image to be corrected, from the pixel value a pixel-specific noise correction factor obtained from a noise reduction model, the noise reduction model comprising a linear component depending on the temperature of the image sensor plus an exponential component depending on the temperature of the image sensor multiplied by an integration time, the linear component and the exponential component depending on pixel-specific coefficients, images are acquired by the image sensor in the absence of light with a minimum integration time, each image being taken at a different respective temperature, and For each pixel in the image to be corrected, a coefficient of the linear component is determined by a linear regression calculation applied to a corresponding pixel in the image captured at the corresponding temperature in the absence of light.

2. The method according to claim 1, wherein The denoising model is defined by the following equation: bm[i,j]=IT×ad[i,j]×Exp(bd[i,j]×TP)+ab[i,j]×TP+bb[i,j] Wherein, bm[i,j] is the noise correction factor to be subtracted from the corresponding pixel of the image to be corrected, IT is the integration time, TP is the temperature of the image sensor, EXP is an exponential function, ad[i,j], bd[i,j], ab[i,j] and bb[i,j] are the pixel-specific coefficients.

3. The method according to claim 1, comprising: The noise correction factor for each pixel is calculated each time the integration time is changed, or each time the temperature of the image sensor deviates from a previous value by more than a temperature deviation threshold.

4. The method according to claim 3, comprising the steps of: images are acquired by the image sensor in the absence of light with different integration times, the image sensor being subjected to different temperatures, generating a correction image obtained by subtracting an image captured by the image sensor in the absence of light at a minimum integration time at the same temperature from each image captured by the image sensor in the absence of light at different integration times, and The coefficients of the exponential components are determined by exponential fit calculations applied to corresponding pixels in correction images obtained at different temperatures and corresponding to the same integration time.

5. The method according to claim 4, wherein The coefficients of the exponential components are determined by averaging the coefficients obtained by exponential fitting calculations for different integration times.

6. The method according to claim 1, wherein The noise reduction model includes the same components for all pixels of the image sensor that depend on the integration time.

7. The method according to claim 1, comprising: A video image stream is obtained, wherein a noise correction factor corresponding to each pixel in an image of the video image stream is subtracted from the corresponding pixel in each image of the video image stream.

8. The method according to claim 1, comprising the steps of: receiving a command to select a gain value for the image sensor, and Based on the selected gain value, a pixel-specific noise correction factor is selected for each pixel in the image to be corrected, the noise correction factor being used to correct the value of each pixel in the image to be corrected, the noise correction factor being determined based on a set of pixel-specific coefficients generated according to the selected gain value.

9. The method according to claim 1, comprising: For each pixel in the corrected image after noise reduction, the value of the pixel is multiplied by a gain correction factor specific to the pixel obtained from the gain normalization table to obtain an image with normalized gain.

10. The method according to claim 9, comprising: The gain normalization table is updated each time the integration time is changed or each time the temperature of the image sensor deviates from a previous value by more than a temperature deviation threshold, the updating of the gain normalization table being implemented by an interpolation calculation applied to a set of gain normalization tables determined for different temperatures.

11. The method according to claim 10, comprising the steps of: The image sensor acquires a sequence of images in the presence of a uniform light source with different integration times or at different intensities of the light source, each sequence of images being taken at a different corresponding temperature. determining, for each pixel in one image in each sequence of images, a gain by a linear regression calculation applied to the corresponding pixel in the images in the sequence of images, and For each pixel in one image in each image sequence, a gain correction factor is determined by dividing the average of the gains obtained for all pixels of the image sequence by the gain determined for the pixel.

12. The method according to claim 1, comprising the steps of: a sequence of images acquired by the image sensor in the absence of light with an average integration time, each image being acquired with the image sensor subjected to a corresponding temperature, calculating, for each pixel in an image in the acquired image sequence, an average of deviations at different image sensor temperatures, each deviation being calculated for an image sensor temperature, the deviation being a deviation between a value of the pixel in the image in the image sequence corresponding to the image sensor temperature and the noise correction factor defined for the pixel at the image sensor temperature with the average integration time; and The average of the deviations is compared to a threshold, and if the average of the deviations is greater than the threshold for a pixel, the pixel is considered defective.

13. The method according to claim 12, comprising the steps of: Correcting each image obtained by the image sensor by replacing the value of a defective pixel with the value of an adjacent pixel or with the average of adjacent pixels, or Each image obtained by the image sensor is corrected by replacing the corresponding values ​​of the defective pixel and the pixels adjacent to the defective pixel with the values ​​of the pixels adjacent to the defective pixel and the pixels adjacent to the defective pixel or the average value of the pixels adjacent to the defective pixel and the pixels adjacent to the defective pixel.

14. A device for correcting defects occurring in an image generated by an image sensor, configured to implement the method according to any one of claims 1 to 13.

15. An imaging device comprising: An image sensor, a circuit for obtaining a temperature of the image sensor, and a circuit for obtaining an integration time applied to the image sensor, the imaging device being configured to implement the method according to any one of claims 1 to 13.

16. The apparatus according to claim 14 or 15, wherein The image sensor is of CTIA or SF type.

17. A computer program product loadable into a memory of a computer, said computer program product, when executed by said computer, configuring said computer to perform the method according to any one of claims 1 to 13.

Images