CN120166312A - A dark current correction method for image sensor - Google Patents

Abstract

The present invention relates to the field of image sensor correction technology, and solves the technical problem of traditional methods in realizing efficient and low-cost dark current correction in high dynamic range, high temperature environment or extremely low illumination scene, and in particular relates to a dark current correction method for an image sensor, including building a dark field image acquisition environment, collecting a dark field image data set; constructing a temperature-long/short exposure-space model; obtaining model calibration parameters; performing real-time correction on the dark field image, and quantifying the correction effect, and comparing the image uniformity before and after correction. The present invention constructs a temperature-long/short exposure-space model by comprehensively considering the relationship between dark current and exposure time, temperature, and pixel position, so as to accurately model the dark current characteristics of each pixel, improve the correction accuracy, and reduce the computational complexity by adjusting the model structure, so that it is suitable for real-time image processing systems, and improves the imaging quality of sensors in complex environments such as long exposure and high temperature.

Description

Dark current correction method of image sensor

Technical Field

The present invention relates to the field of image sensor correction technologies, and in particular, to a dark current correction method for an image sensor.

Background

Dark current refers to the phenomenon that pixels in an image sensor still generate an electronic signal in the absence of light, which signal may appear as additional noise in the image. Therefore, in the long exposure imaging process of the image sensor, the accumulation of dark current can cause signal offset, and the imaging quality is affected.

Aiming at the influence of dark current, the prior art proposes a dark current correction method of a CMOS image sensor with the publication number of CN111432093A, a correction method and a system of dark current of a semiconductor photoelectric sensor with the publication number of CN115734089B, an infrared non-uniform correction method and a system based on dark current fitting with the publication number of CN118714419A and a CCD camera response non-uniformity correction method for realizing dark current compensation with the publication number of CN 110855915A. However, in the prior art, the dark current correction for the image sensor mainly adopts methods of in-pixel real-time calibration, black level reference compensation, dark field image subtraction correction, OTP NVM storage correction, automatic black level compensation and the like.

These methods can reduce Fixed Pattern Noise (FPN), improve signal-to-noise ratio (SNR), optimize long exposure image quality to varying degrees. However, these techniques often have problems such as inaccurate models, limited correction accuracy, etc., and it is difficult to achieve efficient, low-cost dark current correction in a High Dynamic Range (HDR), high temperature environment, or very low illuminance scene.

Disclosure of Invention

Aiming at the defects of the prior art, the invention provides a dark current correction method of an image sensor, which solves the technical problems of high efficiency and low cost of dark current correction in a high dynamic range, a high temperature environment or an extremely low illumination scene in the traditional method.

In order to solve the technical problems, the invention provides a dark current correction method of an image sensor, which comprises the following steps:

S1, building a dark field image acquisition environment, and acquiring a plurality of groups of dark field images at different temperatures T and different exposure times T _exp for averaging to form a dark field image data set D containing a plurality of calibration images D (T _exp, T, x, y);

s2, constructing a pixel level series model of dark signal changing along with exposure time and temperature to obtain a temperature-long/short exposure-space model used for correcting dark current;

s3, obtaining model calibration parameters required by correcting dark current according to the dark field image data set D;

S4, correcting the dark field image in real time based on the model calibration parameters and the temperature-long/short exposure-space model, quantitatively describing the correction effect, and comparing the image uniformity before and after correction.

Further, in step S1, the specific process includes the following steps:

S11, shooting in a full black environment, ensuring no interference of an external light source, covering a lens cover and closing automatic gain on an image sensor to obtain a pure dark current signal;

S12, under each exposure time T _exp and temperature T combination, K Zhang Anchang image D _k is acquired and averaged to obtain a calibration image D (T _exp, T, x, y) under the current exposure time T _exp and temperature T combination, and the formula is as follows:

Wherein D _k represents a dark field image acquired at the kth time;

s13, setting exposure time t _exp, wherein the increase of dark current is different under short exposure and long exposure, so that two types of data including short exposure and long exposure are required to be acquired respectively;

S14, setting the temperature T range as T epsilon { T ₁,T₂,…,T_M };

S15, acquiring calibration images D (T _exp, T, x, y) at different temperatures T and different exposure times T _exp to obtain a dark field image dataset D for acquiring model calibration parameters.

Further, in step S13, collecting two types of data of short exposure and long exposure includes:

short exposure, namely selecting a short-time exposure point to ensure that dark current signals are still in a linear region, wherein the acquisition range is T _i＜t_th, wherein t _th is the critical value for short exposure and long exposure;

Long exposure, in which the accumulation of dark current enters a nonlinear region, a longer time of acquisition of dark field image D _k is required to characterize the growing trend, the acquisition range is t_i≥t_th。

Further, in step S2, the specific process includes the following steps:

s21, converting a dark current-temperature exponential relationship into an exponential form based on e to obtain a temperature dependence formula, wherein:

The expression of the dark current-temperature index relationship is:

The temperature dependence formula is:

Wherein mu _I is dark current, T is current temperature, T _ref is reference temperature when each test is performed in EMVA, mu _I.ref is dark current measured at the reference temperature, and T _d is constant;

s22, establishing a dark current value under short exposure and long exposure at the current temperature T based on a temperature dependency formula

S23, combining dark current value under short exposure and long exposure at the current temperature TA temperature-long/short exposure-space model to correct for dark current was obtained, expressed as:

Wherein, μ _d(x,y,T,t_exp) represents dark current values at different positions, different exposures and different temperatures, μ _d,0 (x, y) represents a fixed value independent of exposure time and temperature, μ _I, ref (x, y) represents a dark current value per exposure time at a reference temperature, α (x, y) represents an exponential increase rate of dark current with temperature, k (x, y) represents a nonlinear saturation rate of dark current at long exposure, and t _th,fixed represents an exposure time boundary between short exposure and long exposure.

Further, in step S22, it includes:

Short exposure when the exposure time T _exp is less than or equal to the critical value T _th, assuming the initial dark signal mu _d,0 as a constant, and taking the temperature dependence formula into the linear increase relation of the dark signal and the exposure time to obtain the current temperature T and the dark current value under short exposure Wherein:

the expression of the dark signal-exposure time linear increase relationship is:

μ_d＝μ_d,0+μ_therm＝μ_d,0+μ_It_exp

Dark current value The expression of (2) is:

Wherein mu _d is a dark signal under the current exposure time, mu _therm is a dark signal of thermally generated electrons, mu _I is a dark current, mu _I,ref is a dark current under the reference temperature, alpha is a temperature sensitivity coefficient, and the dark current is controlled to increase exponentially along with the temperature;

Long exposure, when the exposure time is greater than the critical value t _th, assuming that the relation between the temperature and the dark current under long or short exposure uses a temperature dependency formula, the relation between the exposure time and the dark current under long exposure uses an exponential expression to replace a linear expression, and the continuity at the exposure demarcation is ensured, so that:

wherein k is a long exposure exponential growth rate constant; The current temperature T and the dark current value at a long exposure are shown.

Further, in step S3, the model calibration parameters include an exposure time boundary T _th,fixed, a fixed value μ _d,0 (x, y), a dark current value μ _I,ref (x, y) per unit exposure time at a reference temperature, a reference temperature T _ref, an exponential growth rate α (x, y), and a nonlinear saturation rate k (x, y), wherein:

After the camera reaches the set reference temperature T _ref, acquiring dark field images with different exposures to obtain a change curve of pixel values and exposure time, and obtaining an exposure time boundary T _th,fixed from the segmentation curve;

A fixed value mu _d,0 (x, y) that estimates its initial value from the dark current value measured at the shortest exposure time, i.e., acquires a dark field image of the reference temperature T _ref at the minimum exposure as an initial dark signal, i.e., a fixed value mu _d,0 (x, y);

Dark current value mu _I,ref (x, y) is obtained by fitting dark field images of different exposures in short exposure at reference temperature T _ref;

Reference temperature T _ref, using a standard ambient temperature, or selected according to laboratory conditions;

The exponential growth rate alpha (x, y) is obtained by acquiring dark current fitting measured at different temperatures;

Non-linear saturation rate k (x, y) using dark current values of different exposures in a long exposure at a reference temperature T _ref And (5) calculating to obtain the product.

Further, the specific process of obtaining the dark current value μ _I,ref (x, y) includes:

Obtaining calibration images D (T _exp, T, x, y) of different exposure times at a reference temperature T _ref in the dark-field image dataset D, which are expressed as [ I ₁,I₂,…,I_n ];

Based on the dark signal-exposure time linear increase relationship, a dark current value μ _I,ref (x, y) is calculated for each pixel (x, y) using a least square method, fitting the linear relationship of the exposure time t _exp and the pixel gradation value I (x, y, t _exp), that is:

wherein N is the number of exposure times, i.e. the number of calibration images I _n, I (x, y, t _exp) is the pixel gray value of the calibration image I _n, Σt _exp, The sum and the sum of squares of the exposure times, respectively.

Further, the specific process of obtaining the exponential growth rate α (x, y) includes:

Calculating a dark current sequence mu _I, namely [ mu _I1,μ_I2,…,μ_In ] at different temperatures according to a dark field image dataset D by using a calculation formula of a dark current value mu _I,ref (x, y);

The exponential growth rate α (x, y) is solved based on the formula μ _I＝μ_I,ref·e^(T-Tref)·α, namely:

Taking natural logarithm, namely:

ln(μ_I)＝ln(μ_I,ref)+(T-T_ref)·α

the method is characterized by being arranged into a linear equation set, namely:

ln(μ_I)-ln(μ_I,ref)＝α·(T-T_ref)

let x=t-T _ref、Y＝ln(μ_I)-ln(μ_I,ref), then the formula transforms to y=α·x;

solving by linear regression to obtain α, namely:

Wherein, alpha is the exponential growth rate alpha (x, y).

Further, the specific process of obtaining the nonlinear saturation rate k (x, y) includes:

based on dark current values Let the calculation formula of (1) Then:

Order the I.e.

Since the pixel is close to overexposure in long exposure, the gray level change is relatively slow, i.e. the response of the exposure time to dark current is slowObtaining:

Where k is the nonlinear saturation rate k (x, y).

Further, in step S4, the specific process includes the following steps:

S41, calculating Dark signal images Dark through a temperature-long/short exposure-space model under the current exposure time and temperature;

s42, subtracting the Dark signal image Dark from the input image img of the current frame to obtain a Dark current corrected image imgc;

S43, calculating image uniformity results U ₁ and U ₂ of the input image img and the dark current corrected image imgc, and comparing the image uniformity before and after correction, wherein the calculation formula of the image uniformity is as follows:

In the formula, max (I) is the maximum pixel value of the image, min (I) is the minimum pixel value of the image, mean (I) is the average value of the image, M, N is the row height and the column width of the image respectively;

S44, quantitatively describing the correction effect, namely that the smaller the image uniformity U _i is, the better the image uniformity is, the larger the image uniformity U _i is, the larger the image gray scale difference is, and the worse the image uniformity is;

S45, outputting a dark current corrected image imgc.

By means of the technical scheme, the invention provides a dark current correction method of an image sensor, which has at least the following beneficial effects:

1. The invention constructs a temperature-long/short exposure-space model by comprehensively considering the relation between dark current and exposure time, temperature and pixel position, so as to accurately model the dark current characteristic of each pixel, improve correction precision, reduce calculation complexity by adjusting the model structure, be suitable for a real-time image processing system and improve imaging quality of the sensor in complex environments such as long exposure, high temperature and the like.

2. The temperature-long/short exposure-space model provided by the invention can accurately describe the nonlinear growth characteristic of dark current, reduce the error in a long exposure scene and improve the image quality.

3. According to the invention, single-pixel level correction is realized through fine-granularity modeling, the method is suitable for individual dark current characteristics of different pixels, and the imaging precision is improved.

Drawings

The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute a limitation on the application. In the drawings:

FIG. 1 is a graph showing dark signal of a pixel according to the present invention as a function of exposure time and temperature;

FIG. 2 is a graph drawn from an actual dark field image in accordance with the present invention;

FIG. 3 is a graph showing the average gray level of an image as a function of temperature according to the present invention;

Fig. 4 is a schematic block diagram of a dark current correction system according to the present invention.

Detailed Description

In order that the above-recited objects, features and advantages of the present application will become more readily apparent, a more particular description of the application will be rendered by reference to the appended drawings and appended detailed description. Therefore, the realization process of how to apply the technical means to solve the technical problems and achieve the technical effects can be fully understood and implemented.

Factors influencing dark current in image sensors are temperature (dominant factor), exposure time, spatial heterogeneity, where spatial heterogeneity generally refers to differences in dark current of different pixels due to manufacturing process non-uniformity (e.g., edge pixels are more stressed). The method has the specific effects that as the exposure time increases, dark current signals can be accumulated continuously to influence details of dark parts, dark current is increased exponentially at high temperature to enable the influence of the dark current to be more remarkable, and dark current of different pixels is different to cause that a dark field image presents uneven noise distribution.

Dark current correction is a key step for improving image quality, and the main advantages include reducing Fixed Pattern Noise (FPN), improving signal-to-noise ratio, reducing dark noise, improving long exposure image quality, adapting to different temperature environments, improving imaging stability, improving imaging accuracy of high-precision applications such as science, medical treatment, industry and the like, optimizing subsequent processing accuracy such as AI identification, computer vision and the like, so that dark current correction is an indispensable step regardless of consumer cameras, scientific cameras and industrial imaging. The existing dark current correction method has the following problems:

1. The assumption that dark current grows linearly with exposure time is more accurate in short exposure, but in the case of long exposure, the growing trend shows nonlinear or even exponential changes due to factors such as pixel well filling, carrier recombination, and the like. The existing method cannot effectively characterize the nonlinear effect, so that correction errors in long exposure are increased.

2. Conventional temperature compensation methods generally assume that the entire sensor has uniform temperature response characteristics, failing to model fine-grained for pixel-level temperature variations, resulting in reduced global correction accuracy.

3. Because of the randomness of the manufacturing process, the dark current characteristics of each pixel have individual differences, the existing method mostly adopts global or local statistical average to compensate, and the drift problem of a single pixel level cannot be effectively corrected, and the defects are particularly shown under the requirement of high-precision imaging.

Example 1

Aiming at the defects of the existing method, the embodiment provides a dark current correction method of an image sensor, and a temperature-long/short exposure-space model is constructed by comprehensively considering the relation between dark current and exposure time, temperature and pixel position, so that the dark current characteristics of each pixel are accurately modeled, and the correction precision is improved. Meanwhile, the model structure is adjusted, so that the calculation complexity is reduced, the method is suitable for a real-time image processing system, and the imaging quality of the sensor in complex environments such as long exposure and high temperature is improved. Thereby achieving the effects of improving correction precision and improving real-time processing capacity. The method comprises the following steps:

S1, building a dark field image acquisition environment, and acquiring multiple groups of dark field images at different temperatures T and different exposure times T _exp for averaging to form a dark field image data set D containing multiple calibration images D (T _exp, T, x, y). In order to accurately model the dependency of dark current on exposure time, temperature and pixel position, a full-pixel image covering exposure time T _exp and temperature T is acquired by acquiring a set of dark field images. The specific process comprises the following steps:

S11, shooting in a full black environment, ensuring no interference of an external light source, covering a lens cover and closing automatic gain on an image sensor to obtain a pure dark current signal;

S12, acquiring and averaging K Zhang Anchang images D _k at each combination of exposure time T _exp and temperature T to reduce average noise. Where K is set according to the actual situation, it is sufficient to generally take 5. A calibration image D (T _exp, T, x, y) is obtained at the combination of the current exposure time T _exp and the temperature T. The formula is as follows:

Wherein D _k represents a dark field image acquired at the kth time;

S13, setting exposure time t _exp, wherein the increase of dark current is different under short exposure and long exposure, so that two types of data need to be acquired respectively, namely:

short exposure, namely selecting a short-time exposure point to ensure that dark current signals are still in a linear region, wherein the acquisition range is Where t _th is the critical value of the short exposure and the long exposure, typically between tens of milliseconds and hundreds of milliseconds, can be set according to the actual dark current profile of the camera.

Long exposure, in which the accumulation of dark current enters a nonlinear region, a longer time of acquisition of dark field image D _k is required to characterize the growing trend, the acquisition range is

S14, setting the temperature T range as T epsilon { T ₁,T₂,…,T_M }, and recording the actual temperature during acquisition by using a temperature sensor if an accurate temperature control system is not available.

S15, acquiring calibration images D (T _exp, T, x, y) at different temperatures T and different exposure times T _exp to obtain a dark field image dataset D for acquiring model calibration parameters.

S2, constructing a pixel level series model of dark signal changing along with exposure time and temperature, and obtaining a temperature-long/short exposure-space model used for correcting dark current. EMVA1288 is a standard established by the european machine vision institute (EuropeanMachineVisionAssociation, EMVA) to provide a uniform framework for performance assessment of image sensors and cameras. The standard covers the measurement method of key parameters such as quantum efficiency, dark current, absolute sensitivity threshold and the like, and ensures the comparability and transparency of different device performances. The description of dark current in the standard is largely divided into the following two:

1. The dark signal is not a fixed value, mainly because a part of the dark signal is derived from the contribution of thermally generated electrons, and therefore the dark signal grows linearly with increasing exposure time. The dark signal-exposure time linear increase relationship is described as follows:

μ_d＝μ_d,0+μ_therm＝μ_d,0+μ_It_exp (1)

Wherein μ _d is a dark signal at the current exposure time, μ _therm is a dark signal of thermally generated electrons, μ _d,0 is an initial dark signal, μ _I is a dark current, and t _exp is an exposure time.

2. The temperature dependence of the dark current can be simulated in a simplified form. Because heat generates charge, dark current is approximately exponentially related to temperature increase. The dark current-temperature index relationship is described as follows:

Wherein μ _I.ref is the dark current measured at this reference temperature, T _d is a constant in K or DEG C, which value indicates the temperature interval that causes dark current doubling, i.e. dark current doubling temperature difference, T is the current temperature, and T _ref is the reference temperature at which the tests in EMVA are performed.

However, as the exposure time is longer, the accumulation of dark current can enter the nonlinear region, mainly due to the gradual filling of the pixel wells, charge recombination effects, and other device-level physical processes, resulting in slower acceleration. May be adjusted to an exponential model approximation. In summary, the dark current correction model is optimized by combining the relation between the dark current and the exposure time and temperature, so that the high-efficiency dark current correction can be realized.

The pixel level mathematical model of dark signal variation with exposure time and temperature is constructed by combining the formula (1) and the formula (2) described above, and the specific process comprises the following steps:

s21, converting a dark current-temperature exponential relationship into an exponential form based on e to obtain a temperature dependence formula, wherein:

The expression of the dark current-temperature index relationship is:

The temperature dependence formula is:

Wherein μ _I is a dark current, T is a current temperature, T _ref is a reference temperature at which each test in EMVA is performed, μ _I.ref is a dark current measured at the reference temperature, T _d is a constant in K or DEG C, which indicates a temperature interval at which doubling of the dark current is caused, i.e., a dark current doubling temperature difference;

s22, establishing a dark current value under short exposure and long exposure at the current temperature T based on a temperature dependency formula Namely:

Short exposure when the exposure time T _exp is less than or equal to the critical value T _th, assuming the initial dark signal mu _d,0 as a constant, and taking the temperature dependence formula into the linear increase relation of the dark signal and the exposure time to obtain the current temperature T and the dark current value under short exposure Wherein:

the expression of the dark signal-exposure time linear increase relationship is:

μ_d＝μ_d,0+μ_therm＝μ_d,0+μ_It_exp

Dark current value The expression of (2) is:

Wherein mu _d is a dark signal under the current exposure time, mu _therm is a dark signal of thermally generated electrons, mu _I is a dark current, mu _I,ref is a dark current under the reference temperature, alpha is a temperature sensitivity coefficient, and the dark current is controlled to increase exponentially along with the temperature;

Long exposure, when the exposure time is greater than the critical value t _th, assuming that the relation between the temperature and the dark current under long or short exposure uses a temperature dependency formula, the relation between the exposure time and the dark current under long exposure uses an exponential expression to replace a linear expression, and the continuity at the exposure demarcation is ensured, so that:

wherein k is a long exposure exponential growth rate constant; representing the current temperature T and the dark current value under long exposure;

s23, combining dark current value under short exposure and long exposure at the current temperature T A temperature-long/short exposure-space model to correct for dark current was obtained, expressed as:

Where μ _d(x,y,T,t_exp) represents dark current values at different positions, different exposures, different temperatures, μ _d,0 (x, y) represents fixed noise independent of exposure time and temperature, or assuming that the effect of temperature on fixed noise is small compared to the effect on dark current, approximately a fixed value, μ _I,ref (x, y) represents dark current values per exposure time at a reference temperature (e.g., 0 ℃), α (x, y) represents an exponential increase rate of dark current with temperature, k (x, y) represents a nonlinear saturation rate of dark current at long exposure, t _th,fixed represents an exposure time boundary of short exposure and long exposure, a transition time from linear to nonlinear, a fixed threshold is used, and spatial heterogeneity is not considered.

And on the whole, the temperature-long/short exposure-space model is constructed. As shown in fig. 1, a schematic diagram of dark signal of a coordinate (x, y) pixel with exposure time and temperature changes.

S3, obtaining model calibration parameters required for correcting dark current according to the dark field image data set D, wherein the model calibration parameters comprise an exposure time dividing line T _th,fixed, a fixed value mu _d,0 (x, y), a dark current value mu _I,ref (x, y) of unit exposure time at a reference temperature, the reference temperature T _ref, an exponential growth rate alpha (x, y) and a nonlinear saturation rate k (x, y). The present embodiment fits model calibration parameters in the estimation model according to the dark field image dataset D in combination with the least squares method, wherein:

exposure time dividing line t _th,fixed using exposure time dividing line t _th,fixed according to the actual response characteristics of the camera with exposure time;

After the camera reaches the set reference temperature T _ref, dark field images with different exposures are acquired, a change curve of pixel values and exposure time is obtained, and an exposure time dividing line T _th,fixed can be obtained from the segmentation curve. As shown in fig. 2, a curve is drawn from an actual dark field image, from which the position of the exposure time dividing line t _th,fixed can be clearly reflected.

A fixed value mu _d,0 (x, y) whose initial value can be estimated from the dark current value measured at the shortest exposure time;

The equation (1) shows that μ _d,0 (x, y) is an initial dark signal, and μ _d,0 (x, y) is considered as a fixed value independent of temperature in this embodiment, and by collecting dark field images at different temperatures and under minimum exposure, it is found that the temperature has very little effect on the initial dark signal, and can be ignored, and as shown in fig. 3, the average gray value of the image varies with temperature. The dark field image of the reference temperature T _ref at the minimum exposure is thus acquired as an initial dark signal, i.e. a fixed value μ _d,0 (x, y).

Dark current values mu _I,ref (x, y) per unit exposure time at the reference temperature are obtained by fitting dark field images of different exposures within a short exposure (exposure time is obtained from the shortest to the exposure time dividing line T _th,fixed) at the reference temperature T _ref, and the specific process comprises the following steps:

Obtaining calibration images D (T _exp, T, x, y) of different exposure times at a reference temperature T _ref in the dark-field image dataset D, which are expressed as [ I ₁,I₂,…,I_n ];

Based on the dark signal-exposure time linear increase relationship, a dark current value μ _I,ref (x, y) is calculated for each pixel (x, y) using a least square method, fitting the linear relationship of the exposure time t _exp and the pixel gradation value I (x, y, t _exp), that is:

wherein N is the number of exposure times, i.e. the number of calibration images I _n, I (x, y, t _exp) is the pixel gray value of the calibration image I _n, Σt _exp, The sum and the sum of squares of the exposure times, respectively.

Reference temperature T _ref, using a standard ambient temperature (e.g., 25℃) or selected based on laboratory conditions;

the exponential growth rate alpha (x, y) is generally related to the sensitivity of temperature to dark current, and can be obtained by acquiring dark current fitting measured at different temperatures, and the specific process comprises:

Calculating a dark current sequence mu _I, namely [ mu _I1,μ_I2,…,μ_In ] at different temperatures according to a dark field image dataset D by using a calculation formula of a dark current value mu _I,ref (x, y);

The exponential growth rate α (x, y) is solved based on the formula μ _I＝μ_I,ref·e^(T-Tref)·α, namely:

Taking natural logarithm, namely:

ln(μ_I)＝ln(μ_I,ref)+(T-T_ref)·α

the method is characterized by being arranged into a linear equation set, namely:

ln(μ_I)-ln(μ_I,ref)＝α·(T-T_ref)

let x=t-T _ref、Y＝ln(μ_I)-ln(μ_I,ref), then the formula transforms to y=α·x;

solving by linear regression to obtain α, namely:

Wherein, alpha is the exponential growth rate alpha (x, y);

Non-linear saturation rate k (x, y) using dark current values of different exposures within a long exposure (exposure time from exposure time boundary T _th,fixed to maximum exposure) at reference temperature T _ref And other known parameters provided above, the specific process includes:

based on dark current values Let the calculation formula of (1) Then:

Order the I.e.

Since the pixel is close to overexposure in long exposure, the gray level change is relatively slow, i.e. the response of the exposure time to dark current is slowObtaining:

where k is the nonlinear saturation rate k (x, y), the above calculation is only applicable to cases where k is small, and more accurate methods are needed if the actual k is not small.

S4, correcting the dark field image in real time based on the model calibration parameters and the temperature-long/short exposure-space model, quantitatively describing the correction effect, and comparing the image uniformity before and after correction. The specific process comprises the following steps:

S41, calculating Dark signal images Dark through a temperature-long/short exposure-space model under the current exposure time and temperature;

s42, subtracting the Dark signal image Dark from the input image img of the current frame to obtain a Dark current corrected image imgc;

S43, calculating image uniformity results U ₁ and U ₂ of the input image img and the dark current corrected image imgc, and comparing the image uniformity before and after correction, wherein the calculation formula of the image uniformity is as follows:

wherein:

max (I) is the maximum pixel value of the image, i.e., max (I) =max { I (I, j) |1. Ltoreq.i≤m, 1. Ltoreq.j≤n };

min (I) is the minimum pixel value of the image, i.e., min (I) =min { I (I, j) |1. Ltoreq.i≤m, 1. Ltoreq.j≤n };

mean (I) is the mean of the image, namely:

Wherein M, N is the row and column width of the image respectively;

S44, quantitatively describing the correction effect, namely that the smaller the image uniformity U _i is, the better the image uniformity is, the larger the image uniformity U _i is, the larger the image gray scale difference is, and the worse the image uniformity is;

S45, outputting a dark current corrected image imgc.

Example two

The embodiment provides a dark current correction system for implementing the method based on the dark current correction method provided in the first embodiment, which is composed of a calibration unit, a storage unit and a correction unit, as shown in fig. 4, specifically:

The calibration unit is used for acquiring a sensor calibration image and acquiring a model calibration parameter;

The storage unit is used for storing the calibration parameter file so as to facilitate the subsequent real-time correction processing;

And the correction unit is used for realizing the real-time correction of the dark current according to the model correction parameters acquired by the calibration unit.

In summary, the dark current correction method provided in the embodiment has the following effects:

1. The dark current correction precision is improved, and the traditional method is mostly based on linear or local statistical compensation, so that nonlinear errors under the condition of long exposure cannot be effectively corrected. The invention combines the EMVA1288 existing theory to provide a temperature-long/short exposure-space correction model, which can accurately describe the nonlinear growth characteristic of dark current, reduce the error in a long exposure scene and improve the image quality.

2. Conventional approaches generally assume that the sensor as a whole has the same temperature response, without taking into account pixel-level temperature differences, resulting in a limited global correction accuracy. According to the invention, single-pixel level correction is realized through fine-granularity modeling, the method is suitable for individual dark current characteristics of different pixels, and the imaging precision is improved.

3. The model structure is adjusted, the calculation complexity is reduced, the method is suitable for a real-time image processing system, and the imaging quality of the sensor in complex environments such as long exposure and high temperature is improved.

Those of ordinary skill in the art will appreciate that all or a portion of the steps in a method of implementing an embodiment described above may be implemented by a program to instruct related hardware, and thus, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

In this specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different manner from other embodiments, so that the same or similar parts between the embodiments are referred to each other. For each of the above embodiments, since it is substantially similar to the method embodiment, the description is relatively simple, and reference should be made to the description of the method embodiment for relevant points.

While the foregoing embodiments have been described in some detail to illustrate the principles and embodiments of the invention, it is to be understood that this description is only for purposes of clarity of understanding and understanding the principles and concepts of the invention, and that the invention is not limited to the specific embodiments and applications described herein, as will be apparent to one of ordinary skill in the art in light of the teachings of the invention.

Claims (10)

1. A dark current correction method of an image sensor, the method comprising the steps of:

S1, building a dark field image acquisition environment, and acquiring a plurality of groups of dark field images at different temperatures T and different exposure times T _exp for averaging to form a dark field image data set D containing a plurality of calibration images D (T _exp, T, x, y);

s2, constructing a pixel level series model of dark signal changing along with exposure time and temperature to obtain a temperature-long/short exposure-space model used for correcting dark current;

s3, obtaining model calibration parameters required by correcting dark current according to the dark field image data set D;

S4, correcting the dark field image in real time based on the model calibration parameters and the temperature-long/short exposure-space model, quantitatively describing the correction effect, and comparing the image uniformity before and after correction.

2. The dark current correction method according to claim 1, characterized in that in step S1, the specific process includes the steps of:

S11, shooting in a full black environment, ensuring no interference of an external light source, covering a lens cover and closing automatic gain on an image sensor to obtain a pure dark current signal;

S12, under each exposure time T _exp and temperature T combination, K Zhang Anchang image D _k is acquired and averaged to obtain a calibration image D (T _exp, T, x, y) under the current exposure time T _exp and temperature T combination, and the formula is as follows:

Wherein D _k represents a dark field image acquired at the kth time;

s13, setting exposure time t _exp, wherein the increase of dark current is different under short exposure and long exposure, so that two types of data including short exposure and long exposure are required to be acquired respectively;

S14, setting the temperature T range as T epsilon { T ₁,T₂,…,T_M };

S15, acquiring calibration images D (T _exp, T, x, y) at different temperatures T and different exposure times T _exp to obtain a dark field image dataset D for acquiring model calibration parameters.

3. The dark current correction method according to claim 2, characterized in that in step S13, collecting two types of data of short exposure and long exposure includes:

short exposure, namely selecting a short-time exposure point to ensure that dark current signals are still in a linear region, wherein the acquisition range is Wherein t _th is the critical value for short exposure and long exposure;

Long exposure, in which the accumulation of dark current enters a nonlinear region, a longer time of acquisition of dark field image D _k is required to characterize the growing trend, the acquisition range is

4. A dark current correction method according to claim 3, characterized in that in step S2, the specific procedure comprises the steps of:

s21, converting a dark current-temperature exponential relationship into an exponential form based on e to obtain a temperature dependence formula, wherein:

The expression of the dark current-temperature index relationship is:

The temperature dependence formula is:

Wherein mu _I is dark current, T is current temperature, T _ref is reference temperature when each test is performed in EMVA, mu _I.ref is dark current measured at the reference temperature, and T _d is constant;

s22, establishing a dark current value under short exposure and long exposure at the current temperature T based on a temperature dependency formula

S23, combining dark current value under short exposure and long exposure at the current temperature TA temperature-long/short exposure-space model to correct for dark current was obtained, expressed as:

Wherein, μ _d(x,y,T,t_exp) represents dark current values at different positions, different exposures and different temperatures, μ _d,0 (x, y) represents a fixed value independent of exposure time and temperature, μ _I,ref (x, y) represents a dark current value per unit exposure time at a reference temperature, α (x, y) represents an exponential increase rate of dark current with temperature, k (x, y) represents a nonlinear saturation rate of dark current at long exposure, and t _th,fixed represents an exposure time boundary of short exposure and long exposure.

5. The dark current correction method according to claim 4, characterized in that in step S22, it includes:

Short exposure when the exposure time T _exp is less than or equal to the critical value T _th, assuming the initial dark signal mu _d,0 as a constant, and taking the temperature dependence formula into the linear increase relation of the dark signal and the exposure time to obtain the current temperature T and the dark current value under short exposure Wherein:

the expression of the dark signal-exposure time linear increase relationship is:

μ_d＝μ_d,0+μ_therm＝μ_d,0+μ_It_exp

Dark current value The expression of (2) is:

Wherein mu _d is a dark signal under the current exposure time, mu _therm is a dark signal of thermally generated electrons, mu _I is a dark current, mu _I,ref is a dark current under the reference temperature, alpha is a temperature sensitivity coefficient, and the dark current is controlled to increase exponentially along with the temperature;

Long exposure, when the exposure time is greater than the critical value t _th, assuming that the relation between the temperature and the dark current under long or short exposure uses a temperature dependency formula, the relation between the exposure time and the dark current under long exposure uses an exponential expression to replace a linear expression, and the continuity at the exposure demarcation is ensured, so that:

wherein k is a long exposure exponential growth rate constant; The current temperature T and the dark current value at a long exposure are shown.

6. The dark current correction method according to claim 5, characterized in that in step S3, the model calibration parameters include an exposure time dividing line T _th,fixed, a fixed value μ _d,0 (x, y), a dark current value μ _I,ref (x, y) per unit exposure time at a reference temperature, a reference temperature T _ref, an exponential growth rate α (x, y), and a nonlinear saturation rate k (x, y), wherein:

After the camera reaches the set reference temperature T _ref, acquiring dark field images with different exposures to obtain a change curve of pixel values and exposure time, and obtaining an exposure time boundary T _th,fixed from the segmentation curve;

A fixed value mu _d,0 (x, y) that estimates its initial value from the dark current value measured at the shortest exposure time, i.e., acquires a dark field image of the reference temperature T _ref at the minimum exposure as an initial dark signal, i.e., a fixed value mu _d,0 (x, y);

Dark current value mu _I,ref (x, y) is obtained by fitting dark field images of different exposures in short exposure at reference temperature T _ref;

Reference temperature T _ref, using a standard ambient temperature, or selected according to laboratory conditions;

The exponential growth rate alpha (x, y) is obtained by acquiring dark current fitting measured at different temperatures;

Non-linear saturation rate k (x, y) using dark current values of different exposures in a long exposure at a reference temperature T _ref And (5) calculating to obtain the product.

7. The dark current correction method according to claim 6, characterized in that the specific procedure of obtaining the dark current value μ _I,ref (x, y) includes:

Obtaining calibration images D (T _exp, T, x, y) of different exposure times at a reference temperature T _ref in the dark-field image dataset D, which are expressed as [ I ₁,I₂,…,I_n ];

Based on the dark signal-exposure time linear increase relationship, a dark current value μ _I,ref (x, y) is calculated for each pixel (x, y) using a least square method, fitting the linear relationship of the exposure time t _exp and the pixel gradation value I (x, y, t _exp), that is:

wherein N is the number of exposure times, i.e. the number of calibration images I _n, I (x, y, t _exp) is the pixel gray value of the calibration image I _n, Σt _exp, The sum and the sum of squares of the exposure times, respectively.

8. The dark current correction method according to claim 7, characterized in that the specific procedure of obtaining the exponential growth rate α (x, y) comprises:

Calculating a dark current sequence mu _I, namely [ mu _I1,μ_I2,…,μ_In ] at different temperatures according to a dark field image dataset D by using a calculation formula of a dark current value mu _I,ref (x, y);

The exponential growth rate α (x, y) is solved based on the formula μ _I＝μ_I,ref·e^(T-Tref)·α, namely:

Taking natural logarithm, namely:

ln(μ_I)＝ln(μ_I,ref)+(T-T_ref)·α

the method is characterized by being arranged into a linear equation set, namely:

ln(μ_I)-ln(μ_I,ref)＝α·(T-T_ref)

let x=t-T _ref、Y＝ln(μ_I)-ln(μ_I,ref), then the formula transforms to y=α·x;

solving by linear regression to obtain α, namely:

Wherein, alpha is the exponential growth rate alpha (x, y).

9. The dark current correction method according to claim 7, characterized in that the specific procedure of obtaining the nonlinear saturation rate k (x, y) comprises:

based on dark current values Let the calculation formula of (1) Then:

Order the I.e.

Since the pixel is close to overexposure in long exposure, the gray level change is relatively slow, i.e. the response of the exposure time to dark current is slowObtaining:

Where k is the nonlinear saturation rate k (x, y).

10. The dark current correction method according to claim 9, characterized in that in step S4, the specific process comprises the steps of:

S41, calculating Dark signal images Dark through a temperature-long/short exposure-space model under the current exposure time and temperature;

s42, subtracting the Dark signal image Dark from the input image img of the current frame to obtain a Dark current corrected image imgc;

S43, calculating image uniformity results U ₁ and U ₂ of the input image img and the dark current corrected image imgc, and comparing the image uniformity before and after correction, wherein the calculation formula of the image uniformity is as follows:

Wherein, max (I) is the maximum pixel value of the image, and min (I) is the minimum pixel value of the image;

mean (I) is the average value of the image, M, N is the row height and column width of the image respectively;

S44, quantitatively describing the correction effect, namely that the smaller the image uniformity U _i is, the better the image uniformity is, the larger the image uniformity U _i is, the larger the image gray scale difference is, and the worse the image uniformity is;

S45, outputting a dark current corrected image imgc.