CN102300057A - Method for correcting response inconsistency of linear array CCD (Charge Coupled Device) image elements - Google Patents

Abstract

The invention discloses a method for correcting response inconsistency of linear array CCD (Charge Coupled Device) image elements. According to the characteristics of a CCD, the existing two-point flat field correction method is improved based on correct calculation of dark current noise; and based on formation mechanism of response inconsistency of the image elements, an inconsistency correction method is divided into two independent parts (dark current noise and inconsistency correction and photoelectric response inconsistency correction), and digital gain correction is also introduced, finally, the correction of response inconsistency of the linear array CCD image elements is implemented. The method provided by the invention can be used for overcoming the defect that, in a two-point correction method, values of dark image elements cannot correctly reflect present dark current noise when the CCD is partially saturated; and the method can be widely used in the present imaging devices, has the characteristics of being easy to realize and high in precision, and can be used for eliminating image degradation caused by dark current noise and image element response inconsistency noise and improving the imaging performance of a detector.

Description

The bearing calibration of line array CCD pixel response inconsistency

Technical field

The present invention relates to a kind of bearing calibration of line array CCD pixel response inconsistency.

Background technology

Charge coupled device (Charge-Coupled Devices) is called for short CCD, is a kind of semiconductor imaging device based on photoelectricity transformation principle.CCD is the core image device of present imaging system, and the charge-trapping that it can produce incident photon and is exported collected electric charge by control circuit in its potential well in the mode that electric charge shifts, thereby formation is corresponding to the digital picture of optical imagery.Comprising various noises in the output signal of CCD, as photon shot noise, dark current noise, fixed pattern noise with read noise etc., the existence of these noises has had a strong impact on the detectivity and the spatial resolution of remote sensor, and therefore inhibition or removal of images noise are the effective means that improves the detector image-forming performance.

The desired light electroresponse model of single CCD pixel is:

y _i＝ηx _i+n (1)

Wherein, y _iBe output signal voltage, x _iBe the light intensity of input, η is the optical responsivity of CCD, and n is dark output voltage (comprising some random noises), and i represents i the pixel of CCD.By formula (1) as can be known, the output signal voltage of single CCD pixel satisfies linear relationship with the input light intensity.

In theory, the optical responsivity η of each pixel and dark output voltage n should be in full accord among the CCD, but in fact because the existence of photoelectric respone inconsistency and dark current noise inconsistency, make that the η and the n of different pixels are all inequality, this just causes CCD under even illumination condition, there are some differences in the output of each pixel, and this phenomenon is called pixel response inconsistency.This inconsistency can be described as:

y _i＝η _ix _i+n _i (2)

In following formula, η _iAnd n _iBe respectively the optical responsivity and the dark output voltage of i pixel, wherein once a coefficient has reflected the photoelectric respone inconsistency, and constant term has reflected the inconsistency of dark current noise.

By formula (1) and (2) as can be known, desired light electroresponse model and the realistic model of each pixel of CCD under even light incident condition is as follows respectively:

y＝ηx+n (3)

y _i＝η _ix+n _i (4)

In the formula, x is the even light of input, y and y _iRepresent ideal output and the actual output of CCD respectively.

Above-mentioned two formula simultaneous can be got:

$y=\frac{\mathrm{\η}}{{\mathrm{\η}}_i}{y}_i+n-\frac{\mathrm{\η}}{{\mathrm{\η}}_i}{n}_i=k_i{y}_i+b_i---\left(5\right)$

The inconsistency that existing two point calibration methods adopt following formula to carry out the pixel response is proofreaied and correct:

s _i＝k _iy _i+b _i (6)

Wherein, s _iBe the image after proofreading and correct, Be called the gain factor and the side-play amount of correction.Tradition two point calibration methods are only used two groups of calibration coefficients to finish inconsistency and are proofreaied and correct (comprising that dark current noise and inconsistency thereof are proofreaied and correct and the photoelectric respone inconsistency is proofreaied and correct), do not consider the influence of the gain of A/D conversion chip simultaneously, so calibration result is relatively poor to the correction result.In addition because this method is not used dark pixel among the CCD, therefore proofread and correct also residual some dark current noises in the image of back.

Summary of the invention

Technology of the present invention is dealt with problems and is: overcome the deficiencies in the prior art, a kind of line array CCD pixel response inconsistency bearing calibration simple, that correction accuracy is high that realizes is provided, can be used in the imaging system based on CCD, to eliminate the image information degeneration that pixel response inconsistency noise causes.

Technical solution of the present invention is: the bearing calibration of line array CCD pixel response inconsistency, adopt formula s _i=k _i(y _i-(b _i* DC+c _i* g)) the pixel response inconsistency of CCD is proofreaied and correct, formula p adopted then _i=k * s _iCarry out digital gain adjustment, wherein p to proofreading and correct the result _iBe the adjusted result of digital gain, k is a gain coefficient,

N is the figure place of A/D conversion chip, and max (DC) is the dark current noise maximum of actual measurement, y _iBe the non-uniform image of input, k _iBe photoelectric respone inconsistency correction coefficient, b _iAnd c _iBe dark current noise and inconsistency correction coefficient thereof, DC is the size of dark current noise, and g is the yield value of A/D conversion chip, and DC adopts the method for iteration to ask for according to the operating state of current C CD,

When | m _Dc-DC _J-1|＞T _DcAnd m _Cr＞T _CrThe time, DC _j=DC _J-1

Work as m _Dc-DC _J-1＞T _DcAnd m _Cr＜T _CrThe time, DC _j=DC _J-1+ T _Dc

Work as DC _J-1-m _Dc＞T _DcAnd m _Cr＜T _CrThe time, DC _j=DC _J-1-T _Dc

Under other situations, DC _j=α DC _J-1+ (1-α) m _Dc

In the formula

Be the average of dark pixel, d _JnRepresent the output signal of dark pixel, N represents the number of dark pixel, DC _jAnd DC _J-1The dark current noise size of representing current line and lastrow respectively, DC _jInitial value DC ₁For

α is the weights by the size decision of random noise among the CCD, T _DcBe the threshold value that is used to judge whether dark pixel suddenlys change, m _CrFor having shifted the average of the pixel that from the horizontal transfer register, additionally reads behind all effective pixels, T _CrFor being used to judge whether to exist the judgment threshold of saturated pixel.

The present invention's advantage compared with prior art is:

(1) the inventive method is according to the characteristics of CCD device, promptly can be in the horizontal transfer register when CCD part pixel is saturated residual charge, by judging the size of the pixel that from the horizontal transfer register, migrates out after effective pixel, judge whether CCD exists the fractional saturation phenomenon; Whether there is the fractional saturation phenomenon according to CCD, whether undergo mutation in conjunction with the value of dark pixel simultaneously, adopt diverse ways to calculate dark current noise under different circumstances, the value that has overcome Traditional calculating methods dark pixel when the CCD fractional saturation can't correctly reflect the shortcoming of current dark current noise;

(2) meter of the present invention has adopted a kind of iterative processing method that is similar to low-pass filtering when calculating dark current noise, has significantly reduced the influence of random noise to result of calculation;

(3) the inventive method is improved existing 2 flat field correcting algorithms, formation mechanism according to pixel response inconsistency, the inconsistency correcting algorithm is divided into dark current noise and inconsistency correction and photoelectric respone inconsistency thereof proofreaies and correct two independent parts, correction accuracy height;

(4) the inventive method has been considered the influence of the gain of A/D conversion chip to the correction result at timing, has improved the signal to noise ratio of proofreading and correct the back image; In trimming process, introduce the digital gain link image has been carried out the gray scale stretch processing, solved the problem that view data after proofreading and correct can't reach A/D conversion chip full scale.

Description of drawings

Fig. 1 is the FB(flow block) of the inventive method;

Fig. 2 adopts the inventive method to proofread and correct the dark current noise of front and back under the different integration progression;

Fig. 3 proofreaies and correct the front and back different light intensity hypograph grey scale curve after column direction is averaged respectively for adopting the inventive method;

Fig. 4 proofreaies and correct the different A/D gain condition hypographs in the front and back grey scale curve after column direction is averaged respectively for adopting the inventive method;

Fig. 5 is for adopting desirable photon transfer curve and the measured curve after the inventive method is proofreaied and correct.

Embodiment

The present invention improves two traditional point calibration methods according to the characteristics and the test result of CCD device, and the FB(flow block) of improvement post-equalization method as shown in Figure 1.This method is at first calculated dark current noise according to the work at present state (being the value of saturated judgement pixel and dark pixel) of CCD, according to dark current noise and calibration coefficient view data is carried out correction of dark current noise and inconsistency thereof and the correction of photoelectric respone inconsistency respectively afterwards, by the digital gain link view data after proofreading and correct is carried out gray scale at last and stretch.

1, the calculating of dark current noise

Semiconductor is inner to be transferred under the effect of driving pulse because the charge carrier that warm-up movement produces is filled potential well, and forms electric current at output, even under unglazed fully situation, also exist, i.e. and dark current noise.All ccd sensors all can be subjected to the influence of dark current noise, its existence has limited the sensitivity and the dynamic range of device, and the size of dark current noise and the relation of temperature are very close, 5 ℃～6 ℃ of the every increases of temperature, and dark current noise will be increased to original 2 times.

The present most of CCD that uses, can be manufactured with a small amount of dark pixel at the section start of pixel array (covered does not expose to scenery, but still have dark current noise to produce), by its output signal sampling is stored, and subtract each other with the output signal sampled value of follow-up effective pixel, just can remove dark current noise.But the output signal of dark pixel also can be subjected to the influence of various random noises among the CCD, if therefore directly with the output signal of the dark pixel benchmark as the dark current noise size, will introduce various random noises.The present invention is in order to suppress the influence of random noise, and the average of each dark pixel has been carried out the processing of similar low-pass filtering during to each exposure, is shown below:

${\mathrm{DC}}_j=\left\{\begin{array}{cc}\frac{1}{N}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{d}_{\mathrm{jn}}& j=1\\ \mathrm{\α}{\mathrm{DC}}_{j-1}+(1-\mathrm{\α})\frac{1}{N}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{d}_{\mathrm{jn}}& j\≠1\end{array}\right.---\left(7\right)$

In the formula, d _JnRepresent the output signal of dark pixel, N represents the number of dark pixel, DC _jThe size of expression current line dark current noise, α is that its value utilizes following method to determine by the weights of the size decision of random noise among the CCD.

Because the time interval between the adjacent double exposure of line array CCD is very short, so the dark current noise among the CCD should approximately equal, i.e. DC during adjacent double exposure _j≈ DC _J-1, therefore as can be known by formula (7):

$(1-\mathrm{\α})\frac{1}{N}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{d}_{\mathrm{jn}}\≈(1-\mathrm{\α}){\mathrm{DC}}_{j-1}---\left(8\right)$

Consider the influence of random noise to dark pixel output signal, therefore:

$\frac{1}{N}\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{d}_{\mathrm{jn}}\≈{\mathrm{DC}}_{j-1}+\mathrm{\ϵ}---\left(9\right)$

Substitution formula (7) can get:

DC _j≈αDC _j-1+(1-α)(DC _j-1+ε)＝DC _j-1+(1-α)ε (10)

In the actual measurement, the size of random noise is approximate to satisfy 50≤ε≤100, when calculating for make random noise to the influence of iterative value less than 1 DN value, be convenient to the hardware realization simultaneously, among the present invention

When 2, CCD part pixel is saturated to the influence of dark current noise

Because the transfer efficiency of horizontal transfer register can't accomplish 100% among the CCD, when wherein the part pixel is saturated, can be in the horizontal transfer register residual some electric charges down of these saturated pixels, and in the dark pixel that these residual charges are exported can be superimposed upon imaging next time the time, therefore this moment dark pixel the untrue reflection current C of value CD in dark current noise.If still calculate dark current noise according to formula (7), then the dark current noise that calculates this moment can be greater than dark current noise actual among the CCD, thus the effect correction result.In order to remove the influence to dark pixel when saturated of part pixel, the present invention has carried out following processing:

(1) whether dark pixel undergos mutation: the average m that judges dark pixel _DcWith current iteration value DC _J-1Whether the absolute value of difference is greater than threshold value T _Dc, if greater than showing that then sudden change has taken place dark pixel;

(2) whether there is the saturated phenomenon of part pixel among the CCD: by design CCD drive signal, make and from the horizontal transfer register, additionally read several pixels (being called fractional saturation judgement pixel) after having shifted all effective pixels, judge the average m of these several pixels _CrWhether greater than threshold value T _Cr, if greater than showing that then the part pixel is saturated among the CCD.

When calculating dark current noise, need carry out following processing according to above two conditions:

if|m _dc-DC _j-1|＞T _dc?and?m _cr＞T _cr

DC _j＝DC _j-1

else?if?m _dc-DC _j-1＞T _dc?and?m _cr＜T _cr

DC _j＝DC _j-1+T _dc (11)

else?if?DC _j-1-m _dc＞T _dc?and?m _cr＜T _cr

DC _j＝DC _j-1-T _dc

else

DC _j＝αDC _j-1+(1-α)m _dc

Promptly sudden change has taken place in the output signal when dark pixel, and the part pixel among the CCD is saturated simultaneously, then keeps the iterative value of current dark current noise constant; When sudden change has taken place for the output signal of dark pixel, but when not having part pixel saturated phenomenon among the CCD, then the value of dark current noise equals the current iteration value and adds (subtracting) T _DcIn other cases, calculate dark current noise according to formula (7).

3, inconsistency is proofreaied and correct

By the narration of front as can be known, the pixel of CCD response inconsistency is mainly caused by dark current noise inconsistency and photoelectric respone inconsistency.The dark current noise inconsistency is owing to dark current skewness, each photosensitive first size interval do not wait the spatial distribution noise that causes.Measure as can be known by experiment, the size of the dark current noise of each effective pixel and dark pixel is similar to and satisfies following linear relationship:

d _ci＝b _i×DC+c _i×g (12)

In the formula, d _CiBe i the effectively dark current noise of pixel, DC is the size of dark current noise, and g is the yield value of A/D conversion chip, b _iAnd c _iBe dark current noise and inconsistency correction coefficient thereof.Therefore, dark current noise and inconsistency thereof can be eliminated according to following formula:

z _i＝y _i-(b _i×DC+c _i×g) (13)

In the formula, y _iNon-uniform image for input.

And the photoelectric respone inconsistency is since among the CCD difference of optical responsivity of each pixel cause, adopt flat field correction method to be removed for this part noise, that is:

s _i＝k _iz _i (14)

In the formula, s _iBe the view data after proofreading and correct, k _iBe photoelectric respone inconsistency correction coefficient.

It is as follows that formula (13) and (14) simultaneous can be got updating formula of the present invention:

s _i＝k _i(y _i-(b _i×DC+c _i×g)) (15)

In addition, the photoelectric respone inconsistency correction coefficient k in the top narration _i, and dark current noise and inconsistency coefficient b thereof _iAnd c _iBe referred to as calibration coefficient, their value obtains by least square method according to the view data under the different light intensity of calibration experiment acquisition.

4, digital gain

By formula (15) as can be known, in the implementation process of correcting algorithm, can implement to deduct the dark current noise operation to all effective pixels, this just makes that the view data after proofreading and correct can't reach the full scale of A/D, therefore need carry out gray scale to the view data after proofreading and correct and stretch, digital gain partly is used for finishing this function.The computing formula of this part is as follows:

p _i＝k×s _i (16)

Wherein k is a gain coefficient, and its value is by the maximum decision of the full scale and the dark current noise of A/D conversion chip, that is:

$k=\frac{2^n-1}{\max \left(\mathrm{DC}\right)}---\left(17\right)$

In the formula, n is the figure place of A/D conversion chip, and the maximum of max (DC) expression dark current noise obtains by actual measurement.

Embodiment

Utilize FPGA to realize method of the present invention, and on certain imager circuit, test, the CCD that this imager uses is a TDICCD, the photoelectric respone inconsistency that provides in the device handbook and the representative value of dark current noise inconsistency are 5%, and the maximum of photoelectric respone inconsistency is 10%.For simply, objectively result images is estimated, adopt grey scale curve, photon transfer curve and three indexs of signal to noise ratio (snr) before and after proofreading and correct to compare here.

Fig. 2 has provided dark current noise (the CCD imaging under full secret note part before and after proofreading and correct under the different integration progression, image after the imaging is averaged at column direction respectively), be followed successively by from top to bottom among Fig. 2 (a) integration progression 5, integration progression 4 ..., integration progression 1.As can be seen, the dark current noise of proofreading and correct the back image is significantly less than proofreaies and correct preceding noise from Fig. 2 (b), and the influence of dark current noise and inconsistency thereof has been eliminated in this explanation substantially by bearing calibration of the present invention.Because the dark current noise of proofreading and correct in the low fourth gear integration progression image in back approaches 0, therefore their curve overlaps mutually in Fig. 2 (b).

Fig. 3 (a) has provided and has proofreaied and correct the preceding different light intensity hypograph grey scale curve after column direction is averaged respectively, and Fig. 3 (b) is the grey scale curve after proofreading and correct.Therefrom scheme as can be seen, the inconsistency of proofreading and correct the back image has also obtained tangible improvement.

Fig. 4 (a) has provided and has proofreaied and correct the preceding different analog-digital chip gain condition hypographs grey scale curve after column direction is averaged respectively, is followed successively by 9dB, 6dB and 0dB from top to bottom, and Fig. 4 (b) is the grey scale curve after proofreading and correct.Therefrom scheme as can be seen, the inconsistency of the image after the correction under each gain condition all obtained obvious improvement.

The photon transfer curve of CCD has reflected the relation of each noise (standard deviation of output signal) with incident intensity, usually under different input light intensity, test and obtain, draw with logarithmic curve for the whole dynamic range that covers CCD, desirable photon transfer curve is shown in Fig. 5 (a).As seen from the figure, it is made up of three different zones: the relation of reading noise (comprising dark current noise and thermal noise etc.) and input signal, the standard deviation sigma of reading noise have been reflected in first zone _sBe to record under complete black condition, the intensity of this part noise and input signal is irrelevant; Along with the increase of input signal strength, the photon shot noise plays a major role, because the equivalent charge number of photon shot noise is directly proportional with the square root of incident light subnumber, so slope of a curve is 1/2 in second zone; The 3rd zone is relevant with fixed pattern noise, i.e. photoelectric respone inconsistency, and it is linear with the input light intensity, so the slope of this part is 1.Fig. 5 (b) has provided measured curve, and the even light of exporting with integrating sphere during test is as input.The curve of star has provided the photon transfer curve of entire image among the figure, therefrom can it is evident that 3 zoness of different narrating above; The curve that point constitutes is that the average according to each row standard deviation in the image obtains, and it is equivalent to the curve when not having fixed pattern noise, so this curve is only formed by preceding two sections of curve among Fig. 5 (a).Fig. 5 (c) has provided the ideal curve after proofreading and correct, and proofread and correct owing to carried out pixel response inconsistency, so this curve only is made up of two parts, and not having slope is 1 part.Fig. 5 (d) has provided imager and has carried out measured curve after inconsistency is proofreaied and correct, wherein the curve of star has provided the photon transfer curve of entire image, and the curve that connects and composes of point is obtained by the average of each row standard deviation in the image, two curves overlap substantially among the figure, and the influence of having eliminated pixel response inconsistency by calibration model of the present invention substantially is described.

Table 1 has provided different AD gain (0dB, 6B and 9dB) different DN values signal to noise ratio of the preceding image of (percentage of full scale) correction down, because what use in the imager is line array CCD, therefore the signal to noise ratio of image column direction is apparently higher than the signal to noise ratio of line direction and entire image before proofreading and correct, and the DN value is high more, the gap of line direction signal to noise ratio and column direction signal to noise ratio is obvious more, and this has illustrated that pixel response inconsistency noise plays a leading role when high brightness.Table 2 has provided same lagged product and has divided sum of series to proofread and correct the signal to noise ratio of back image the time of integration down, as can be seen from the table, the signal to noise ratio of proofreading and correct back line direction and entire image has had and has significantly improved, very approaching with the signal to noise ratio of column direction, illustrate that correcting algorithm of the present invention removed pixel response inconsistency noise preferably.

The gain signal to noise ratio of image before the correction under the different DN values of the different AD of table 1

The gain signal to noise ratio of image after the correction under the different DN values of the different AD of table 2

The content that is not described in detail in the specification of the present invention belongs to those skilled in the art's known technology.

Claims (1)

1. line array CCD pixel response inconsistency bearing calibration is characterized in that: adopt formula s _i=k _i(y _i-(b _i* DC+c _i* g)) the pixel response inconsistency of CCD is proofreaied and correct, formula p adopted then _i=k * s _iCarry out digital gain adjustment, wherein p to proofreading and correct the result _iBe the adjusted result of digital gain, k is a gain coefficient,

N is the figure place of A/D conversion chip, and max (DC) is the dark current noise maximum of actual measurement, y _iBe the non-uniform image of input, k _iBe photoelectric respone inconsistency correction coefficient, b _iAnd c _iBe dark current noise and inconsistency correction coefficient thereof, DC is the size of dark current noise, and g is the yield value of A/D conversion chip, and DC adopts the method for iteration to ask for according to the operating state of current C CD,

When | m _Dc-DC _J-1|＞T _DcAnd m _Cr＞T _CrThe time, DC _j=DC _J-1

Work as m _Dc-DC _J-1＞T _DcAnd m _Cr＜T _CrThe time, DC _j=DC _J-1+ T _Dc

Work as DC _J-1-m _Dc＞T _DcAnd m _Cr＜T _CrThe time, DC _j=DC _J-1-T _Dc

Under other situations, DC _j=α DC _J-1+ (1-α) m _Dc

In the formula

Be the average of dark pixel, d _JnRepresent the output signal of dark pixel, N represents the number of dark pixel, DC _jAnd DC _J-1The dark current noise size of representing current line and lastrow respectively, DC _jInitial value DC ₁For α is the weights by the size decision of random noise among the CCD, T _DcBe the threshold value that is used to judge whether dark pixel suddenlys change, m _CrFor having shifted the average of the pixel that from the horizontal transfer register, additionally reads behind all effective pixels, T _CrFor being used to judge whether to exist the judgment threshold of saturated pixel.

Images