CN105681693B - TDI-CMOS image sensor FPN bearing calibration - Google Patents
TDI-CMOS image sensor FPN bearing calibration Download PDFInfo
- Publication number
- CN105681693B CN105681693B CN201511034506.5A CN201511034506A CN105681693B CN 105681693 B CN105681693 B CN 105681693B CN 201511034506 A CN201511034506 A CN 201511034506A CN 105681693 B CN105681693 B CN 105681693B
- Authority
- CN
- China
- Prior art keywords
- formula
- vector
- image
- gray value
- tdi
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N25/00—Circuitry of solid-state image sensors [SSIS]; Control thereof
- H04N25/60—Noise processing, e.g. detecting, correcting, reducing or removing noise
- H04N25/67—Noise processing, e.g. detecting, correcting, reducing or removing noise applied to fixed-pattern noise, e.g. non-uniformity of response
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to field of image processings, fringes noise in the removal image to realize versatility, especially realize the horizontal stripe and nicking for effectively eliminating weave in TDI-CMOS image sensor output image.The technical solution adopted by the present invention is that, TDI-CMOS image sensor FPN bearing calibration, assuming that FPN and random noise are all additive noises, and the two is uncorrelated, for the given TDI-CMOS image sensor chip for having manufactured completion, assuming that the corresponding gray value of its FPN is Gaussian Profile equal, that random noise obedience mean value is zero under identical working environment；Firstly the need of capturing sample image data, then successively RFPN and CFPN are estimated and corrected, finally use compensation method, RFPN and CFPN value that estimation obtains are brought into formula (1), i.e., effectively obtain ideal Pixel Information.Present invention is mainly applied to image procossings.
Description
Technical field
The present invention relates to field of image processings, more particularly to (time delay integration type complementary metal aoxidizes to TDI-CMOS
Object semiconductor) FPN (fixed pattern noise) in picture captured by imaging sensor handled.
Background technique
In based on the cumulative TDI-CMOS image sensor realized of analog domain.It is posted due to existing in simulation accumulator circuit
Raw resistance and capacitor, circuit mismatch will lead to output image in TDI (time delay integration) scanning direction (i.e. the direction " along rail ")
Brightness irregularities, and be in periodic damping, outstanding behaviours is periodical horizontal stripe, which is known as row FPN, and (RFPN, row are fixed
Modal noise).And the system structure of sensor column parallel read-out circuit (simulation accumulator and ADC (analog-digital converter) etc.) due to
It is easy to appear mismatches between the column and the column for process deviation, so as to cause output image in the direction vertical with the scanning direction TDI
(i.e. the direction " across rail ") brightness irregularities show as the nicking of light and shade variation, which is known as column FPN, and (CFPN arranges row and fixes
Modal noise).In the output image of TDI-CMOS image sensor, the horizontal stripe as caused by row FPN and as column FPN caused by
Nicking exists simultaneously and weave in, such as Fig. 2, has seriously affected picture quality, and increases the difficulty of FPN correction.
The row FPN of TDI-CMOS image sensor is caused by simulating the specific circuit structure of accumulator as it, is a kind of
Special FPN.Moreover, for the TDI-CMOS image sensor chip for having manufactured completion, after FPN can only be by image
Processing is to correct.Currently, there is much the technology of fringes noise in existing removal image, but these technologies be all only applicable to it is specific
Imaging sensor or data type, such as infrared imaging, high light spectrum image-forming/multispectral imaging, Moderate Imaging Spectroradiomete；
And these technologies can only handle the striped on horizontal or vertical single direction, can not effectively eliminate TDI-CMOS image sensor
Export the horizontal stripe and nicking of weave in image.
Summary of the invention
In order to overcome the deficiencies of the prior art, it realizes fringes noise in the removal image of versatility, especially realizes and effectively eliminate
TDI-CMOS image sensor exports the horizontal stripe and nicking of weave in image.The technical solution adopted by the present invention
It is TDI-CMOS image sensor FPN bearing calibration, it is assumed that FPN and random noise are all additive noises, and the two not phase
It closes, for the given TDI-CMOS image sensor chip for having manufactured completion, it is assumed that the corresponding gray value of its FPN is in identical work
Make to be equal under environment, random noise obeys the Gaussian Profile that mean value is zero；
The pixel array sized of TDI-CMOS image sensor is M × N, i.e. M row N column distinguish row serial number and column serial number
Be denoted as i and j, it is assumed that the photoresponse of different pixels be it is incoherent, then the gray value y (i, j) for exporting pixel (i, j) in image can
To be expressed as formula (1):
Y (i, j)=x (i, j)-a (i)+b (j)+r (i, j) 1≤i≤L, 1≤j≤N (1)
Wherein, L is image in the size on rail direction, and x (i, j) is the desired gray level value of pixel (i, j), and a (i) is
The corresponding RFPN gray value of i row pixel, b (j) are that jth column pixel corresponds to CFPN gray value, and r (i, j) is all in addition to FPN
The corresponding gray value of noise, according to TDI working principle, a (i) meets formula (2).
A (i)=a (i+T) (2)
T=M_TDI+1 (3)
Wherein, M_TDI is the cumulative series of TDI；Firstly the need of capturing sample image data, then successively to RFPN and CFPN
Estimated and corrected, finally use compensation method, RFPN and CFPN value that estimation obtains are brought into formula (1), i.e., it is effective to obtain
To ideal Pixel Information.
Capturing sample image data comprise the concrete steps that, it is assumed that the resolution ratio of TDI-CMOS image sensor is 8-bit, in phase
Acquisition K width average gray value is about 127 under same test environment, i.e., half saturated uniform light image is denoted as Y1,Y2,…,YK,
This K width image is used as to the sample data of estimation FPN gray value, for appointing piece image Yk, formula (1) is expressed as formula (4):
Yk(i, j)=Xk(i,j)-a(i)+b(j)+rk(i,j) 1≤k≤K (4)
Yk(i, j) is pixel (i, j) image, and in sample image collection process, the foundation for choosing semi-saturation gray value is
" EMVAStandard 1288 " standard.
RFPN is eliminated
Image YkRow mean vector be defined as Uk, the average gray value of the i-th row pixel is defined as Uk(i), institute is detected first
Have along the position of first band in rail direction in sample image, and the operation be known as " first trip detection ":
Image YkRow mean vector in the ratio of two neighboring element be defined as Dk(i), as shown in formula (5).
1≤i≤L-1 (5)
Therefore, UkPosition in curve where jump passes through Dk(i) threshold test obtains, i.e., each period of change is most
The row serial number of a line can be obtained by detection afterwards, and the row serial number of the last line of first period of change is denoted as Sk, scheming
As YkIn, from (Sk+ 1) row starts the secondary new image Y of t periodical configuration of interception oneC,k, by YC,kSize be defined as Lc× N,
By formula (2-1) it is found that LcMeet formula (2-6):
Lc=t × T=t × (M+1) (6)
Aforesaid operations are repeated to all K width sample images being collected into, the new image of K width can be constructed altogether, be denoted as
YC,1,YC,2,…,YC,K, at this point, all corresponding horizontal stripes are all in identical position in K width new images, by image YC,kRow
Mean vector is defined as UC,k, then all new images YC,1,YC,2,…,YC,KHead office mean vector UCIt can be expressed as formula (7):
Based on UCEstimate the corresponding RFPN gray value a (i) of the i-th row pixel, 1≤i≤Lc, a (i) expression formula such as formula (8),
A (i)=round (UC(1)-UC(i)) 2≤i≤M+1 (8)
According to formula (1), RFPN is corrected plus the estimated value of the RFPN gray value of corresponding row by pixel original gray value,
In view of the intensity value ranges of 8-bit image are (0-255), it is also necessary to some constraints are carried out to correction course, such as formula (9) institute
Show.
Wherein, z (i, j) is the gray value after pixel (i, j) correction RFPN；
According to formula (1) and formula (9), the gray value z (i, j) after pixel (i, j) correction RFPN is expressed as formula (10):
Z (i, j)=x (i, j)+b (j)+r (i, j) 1≤i≤L, 1≤j≤N (10).
CFPN is eliminated
To all sample image Y1,Y2,…,YKThe new image of K width is obtained after carrying out RFPN correction, is denoted as Z1,Z2,…,ZK。
ZkThe column mean vector of (1≤k≤K) is defined as Vk, then total column mean vector V of all K width new images can be expressed as formula
(11):
Total column mean vector V is averaged by multiple repairing weld to be calculated, and eliminates the influence of random noise, therefore can be with
The corresponding CFPN gray value b (j) (1≤j≤N) of jth column pixel is estimated based on V,
In order to construct the smooth ideal column mean of a curve approximation first based on vector V estimation CFPN gray value b (j)
Vector VC, then vector V subtracts vector VCObtained difference is the CFPN vector estimated；
Vector VCIt can construct, i.e., be filtered using arest neighbors mean value in such a way that adjacent elements several in vector V are averaged
Wave device (Nearest Neighbor Averaging Filter, NNAF) is constructed, the NNAF with parameter, such as formula (12)
It is shown, to construct vector VC:
Wherein, W is variable, VWIt is the column mean vector constructed using the NNAF of design, and VWIt (j) is the flat of jth column pixel
Equal gray value；
Based on VWThe CFPN vector of estimation is defined as BW, and the average gray value of jth column pixel is defined as BW(j), then vector
BWIt can be calculated by formula (13):
BW=floor (V-VW) (13)
Vector BWThe average value of middle all elements is defined as BM,W, and BM,WAbsolute value be defined as BMA,W, such as formula (14) and formula
(15) shown in:
BMA,W=| BM,W|(15)
When N is sufficiently large, the element in CFPN vector is obeyed the Gaussian Profile that mean value is zero and is therefore estimated
The average gray value of all elements should be zero or be approximately zero in CFPN vector, then corresponds to BMA,WIt is worth the smallest BWVector is pair
The best estimate of CFPN vector, by best estimate vector BWIt is denoted as BS, therefore formula 16 can be obtained:
B (i)=BS (16)。
The features of the present invention and beneficial effect are:
The present invention is effectively eliminated according to TDI-CMOS image sensor FPN producing cause using gray value compensation method
TDI-CMOS image sensor distinctive row FPN and column FPN.Intuitively from human eye, as shown in fig. 7, effect is obvious.Through
After algorithm correction, row average value standard deviation has significant improvement.
Detailed description of the invention:
Fig. 1 TDI-CMOS image sensor structural block diagram；
Fig. 2 original uniform light image；
Fig. 3 average gray value Uk(i) with the relationship of row serial number i；
The relationship of the estimated value a (i) and row serial number i of the RFPN gray value of Fig. 4 a cycle；
Fig. 5 column mean vector V and VCCurve；
The relationship of the CFPN gray value b (s) and column serial number s of Fig. 6 estimation；
Fig. 7 measuring image FPN correction front and back vision Contrast on effect: before (a) correcting, after (b) correction；
The row Mean curve comparison of image in Fig. 8 Fig. 7: (a) primitive curve, (b) calibration curve.
Specific embodiment
The present invention is based on the working principle of TDI-CMOS image sensor, pass through the analysis source FPN, noise behavior and its right
The influence of image quality proposes the noise model of TDI-CMOS image sensor, and is based on according to modelling one kind
The FPN bearing calibration of gray value compensation.
Assuming that FPN and random noise are all additive noises, and the two is uncorrelated.Completion has been manufactured for given
TDI-CMOS image sensor chip, it is assumed that the corresponding gray value of its FPN is equal under identical working environment.Random noise
Obey the Gaussian Profile that mean value is zero.
As shown in Figure 1, the pixel array sized of TDI-CMOS image sensor is M × N, i.e. M row N column.By row serial number and
Column serial number is denoted as i and j respectively.Assuming that the photoresponse of different pixels be it is incoherent, then export the gray scale of pixel (i, j) in image
Value y (i, j) can be expressed as formula (1).
Y (i, j)=x (i, j)-a (i)+b (j)+r (i, j) 1≤i≤L, 1≤j≤N (1)
Wherein, L is image in the size on rail direction, and x (i, j) is the desired gray level value of pixel (i, j), and a (i) is
The corresponding RFPN gray value of i row pixel, b (j) are that jth column pixel corresponds to CFPN gray value, and r (i, j) is all in addition to FPN
The corresponding gray value of noise.According to TDI working principle, a (i) meets formula (2).
A (i)=a (i+T) (2)
T=M_TDI+1 (3)
Wherein, M_TDI is the cumulative series of TDI.
A kind of FPN bearing calibration based on gray value compensation proposed by the present invention, it is necessary first to capturing sample image data,
Then successively RFPN and CFPN are estimated and is corrected.
In order to guarantee validity and accuracy to FPN estimated result, need to acquire using image capturing system a large amount of
Image data.It is assumed that the resolution ratio of TDI-CMOS image sensor is 8-bit.It is average that K width is acquired under identical test environment
Gray value is about the uniform light image of 127 (i.e. semi-saturations), is denoted as Y1,Y2,…,YK, this K width image is used as estimation FPN ash
The sample data of angle value.For appointing piece image Yk, formula (1) can be expressed as formula (4).
Yk(i, j)=Xk(i,j)-a(i)+b(j)+rk(i,j)1≤k≤K(4)
In sample image collection process, the foundation for choosing semi-saturation gray value is " EMVA Standard 1288 " standard.
The standard is formulated by European Machine Vision Association (European Machine Vision Association, EMVA), for retouching
State the characteristic parameter of imaging sensor and camera.In the standard, the measurement of imaging sensor spatial non-uniformity is all based on
The image of 50% saturation.Since RFPN and CFPN are a kind of inhomogeneities spatially, which is suitable for TDI-
Cmos image sensor.
(1) RFPN is eliminated
Image YkRow mean vector be defined as Uk, the average gray value of the i-th row pixel is defined as Uk(i).As shown in figure 3,
Due to the presence of RFPN, UkCurve fluctuation it is very big, and jumped in the position where horizontal stripe.However, all sample graphs
The position of horizontal stripe is not fully identical as in.Therefore, when calculating head office's mean vector of all sample images, by direct
It is being averaged the result is that invalid, it is impossible to be used in estimation RFPN.It can be used for estimating that RFPN's is effective total in order to obtain
Row mean vector should be detected along the position of first band in rail direction in all sample images first, and the operation is known as
" first trip detection ".
Image YkRow mean vector in the ratio of two neighboring element be defined as Dk(i), as shown in formula (5).
1≤i≤L-1 (5)
Therefore, UkPosition in curve where jump can easily pass through Dk(i) threshold test obtains, i.e., each
The row serial number of the last line of period of change can be obtained by detection.By the row serial number of the last line of first period of change
It is denoted as Sk.In image YkIn, from (Sk+ 1) row starts the secondary new image Y of t periodical configuration of interception oneC,k.By YC,kSize it is fixed
Justice is Lc× N, by formula (2-1) it is found that LcMeet formula (2-6).
Lc=t × T=t × (M+1) (6)
Aforesaid operations are repeated to all K width sample images being collected into, the new image of K width can be constructed altogether, be denoted as
YC,1,YC,2,…,YC,K.At this point, all corresponding horizontal stripes are all in identical position in K width new images.By image YC,kRow
Mean vector is defined as UC,k, then all new images YC,1,YC,2,…,YC,KHead office mean vector UCIt can be expressed as formula (7).
Head office mean vector UCIt is averaged and is calculated by multiple repairing weld, eliminate the influence of CFPN and random noise,
Therefore U can be based onCEstimate corresponding RFPN gray value a (i) (1≤i≤L of the i-th row pixelc).And according to analysis above, a
(i) and UCIn the variation of element all there is periodicity, then only need to estimate the RFPN gray value of a cycle.Cause
This, a (i) expression formula such as formula (8).A (i) and i relationship are as shown in Figure 4.
A (i)=round (UC(1)-UC(i)) 2≤i≤M+1 (8)
According to formula (1), the estimated value of the RFPN gray value of corresponding row can be added by pixel original gray value to correct
RFPN.Before RFPN correction, need to carry out " first trip detection " operation first.Furthermore, it is contemplated that the gray value model of 8-bit image
Enclose is (0-255), it is also necessary to some constraints is carried out to correction course, as shown in formula (9).
Wherein, z (i, j) is
Pixel (i, j) corrects the gray value after RFPN.
According to formula (1) and formula (9), the gray value z (i, j) after pixel (i, j) correction RFPN can be expressed as formula (10).
Z (i, j)=x (i, j)+b (j)+r (i, j) 1≤i≤L, 1≤j≤N (10)
(2) CFPN is eliminated
To all sample image Y1,Y2,…,YKThe new image of K width is obtained after carrying out RFPN correction, is denoted as Z1,Z2,…,ZK。
ZkThe column mean vector of (1≤k≤K) is defined as Vk, then total column mean vector V of all K width new images can be expressed as formula
(11)。
Total column mean vector V is averaged by multiple repairing weld to be calculated, and eliminates the influence of random noise, therefore can be with
Based on the corresponding CFPN gray value b (j) (1≤j≤N) of V estimation jth column pixel.Due to the presence of CFPN, the curve fluctuation of V is very
Greatly, as shown in lighter curve in Fig. 5.
In order to construct the smooth ideal column mean of a curve approximation first based on vector V estimation CFPN gray value b (j)
Vector VC, as shown in darker curve in Fig. 6.Then vector V subtracts vector VCObtained difference is the CFPN vector estimated.
Vector VCIt can construct, i.e., be filtered using arest neighbors mean value in such a way that adjacent elements several in vector V are averaged
Wave device (Nearest Neighbor Averaging Filter, NNAF) is constructed.In order to construct vector VC, design herein
A kind of NNAF with parameter, as shown in formula (12).
Wherein, W is variable, VWIt is the column mean vector constructed using the NNAF of design, and VWIt (j) is the flat of jth column pixel
Equal gray value.
Based on VWThe CFPN vector of estimation is defined as BW, and the average gray value of jth column pixel is defined as BW(j).Then vector
BWIt can be calculated by formula (13).
BW=floor (V-VW) (13)
Vector BWThe average value of middle all elements is defined as BM,W, and BM,WAbsolute value be defined as BMA,W, such as formula (14) and formula
(15) shown in.
BMA,W=| BM,W| (15) when N is sufficiently large, the element in CFPN vector obeys the Gaussian Profile that mean value is zero.
Therefore, the average gray value of all elements should be zero or be approximately zero in the CFPN vector estimated.Then correspond to BMA,WValue
The smallest BWVector is the best estimate to CFPN vector, by best estimate vector BWIt is denoted as BS.Therefore formula 16 can be obtained.
B (i)=BS (16)
It is calculated as described above, RFPN and CFPN passes through mathematical formulae, using compensation method, estimation is obtained
RFPN and CFPN value are brought into formula (1), can effectively obtain ideal Pixel Information.
Below with reference to a specific example, the present invention is further illustrated.
It is primarily based on the cumulative TDI-CMOS image sensor of 128 grades of analog domains and devises a set of imaging system, use design
Imaging system acquire the uniform light figure that 100 width average gray values are about 127 (i.e. semi-saturations) under identical test environment
Picture.Be then based on formula (5-9), detected by first trip, intercept the image of 3 period sizes, using this image seek row mean value to
Amount, according to formula (8), taking a (1) is 0, obtains image RFPN.Wherein, first trip detection threshold value is set as 1.15.Similarly, above-mentioned sample is utilized
This, obtains image CFPN according to formula (11-16).The above operation passes through MATLAB software realization.Finally respectively under uniform illumination
The image and actual test image of shooting carry out FPN correction.
The experimental results showed that the row average value standard deviation of the image shot under uniform illumination subtracts from 5.6798 LSB before correction
It is small arrived correction after 0.4214 LSB, column mean standard deviation from the 15.2080LSB before correction be reduced to correction after
13.4623 LSB。
From Fig. 7 (a) it can also be seen that primitive curve fluctuation is very big, and there are periodical jumping (at oval marks).It uses
After the method correction FPN of proposition, all jumps are completely disappeared, and as shown in Fig. 7 (b), therefore calibration curve is more smooth than primitive curve
Many.In addition, the shape and tendency of calibration curve are all consistent with primitive curve, show the method proposed in the same of correction FPN
When can retain the detailed information of original image well.
Claims (2)
1. a kind of TDI-CMOS image sensor FPN bearing calibration, characterized in that assuming that FPN and random noise are all that additivity is made an uproar
Sound, and the two is uncorrelated, for the given TDI-CMOS image sensor chip for having manufactured completion, it is assumed that its FPN is corresponding
Gray value be under identical working environment it is equal, random noise obey mean value be zero Gaussian Profile；
The pixel array sized of TDI-CMOS image sensor is M × N, i.e. row serial number and column serial number are denoted as i by M row N column respectively
And j, it is assumed that the photoresponse of different pixels be it is incoherent, then the gray value y (i, j) for exporting pixel (i, j) in image can be with table
It is shown as formula (1):
Y (i, j)=x (i, j)-a (i)+b (j)+r (i, j) 1≤i≤L, 1≤j≤N (1)
Wherein, L is image in the size on rail direction, and x (i, j) is the desired gray level value of pixel (i, j), and a (i) is the i-th row
The corresponding RFPN gray value of pixel, b (j) are that jth column pixel corresponds to CFPN gray value, and r (i, j) is that all in addition to FPN make an uproar
The corresponding gray value of sound, according to TDI working principle, a (i) meets formula (2)
A (i)=a (i+T) (2)
T=M_TDI+1 (3)
Wherein, M_TDI is the cumulative series of TDI；Firstly the need of capturing sample image data, then successively RFPN and CFPN is carried out
Estimation and correction, finally use compensation method, and RFPN and CFPN value that estimation obtains are brought into formula (1), i.e., effectively managed
The Pixel Information thought；
RFPN is eliminated and is comprised the concrete steps that, image YkRow mean vector be defined as Uk, the average gray value of the i-th row pixel is defined as
Uk(i), it detects in all sample images and is detected along the position of first band in rail direction i.e. first trip first:
Image YkRow mean vector in the ratio of two neighboring element be defined as Dk(i), as shown in formula (5):
Therefore, UkPosition in curve where jump passes through Dk(i) threshold test obtains, i.e., each period of change last
Capable row serial number can be obtained by detection, and the row serial number of the last line of first period of change is denoted as Sk, in image Yk
In, from (Sk+ 1) row starts the secondary new image Y of t periodical configuration of interception oneC,k, by YC,kSize be defined as Lc× N, by formula
(2-1) is it is found that LcMeet formula (2-6):
Lc=t × T=t × (M+1) (6)
Aforesaid operations are repeated to all K width sample images being collected into, the new image of K width can be constructed altogether, be denoted as YC,1,
YC,2,…,YC,K, at this point, all corresponding horizontal stripes are all in identical position in K width new images, by image YC,kRow mean value
Vector is defined as UC,k, then all new images YC,1,YC,2,…,YC,KHead office mean vector UCIt can be expressed as formula (7):
Based on UCEstimate corresponding RFPN gray value a (i) (1≤i≤L of the i-th row pixelc), a (i) expression formula such as formula (8),
A (i)=round (UC(1)-UC(i)) 2≤i≤M+1(8)
According to formula (1), the estimated value of the RFPN gray value of corresponding row is added to correct RFPN by pixel original gray value, is considered
Intensity value ranges to 8-bit image are (0-255), it is also necessary to some constraints are carried out to correction course, as shown in formula (9):
Wherein, z (i, j) is the gray value after pixel (i, j) correction RFPN；
According to formula (1) and formula (9), the gray value z (i, j) after pixel (i, j) correction RFPN is expressed as formula (10):
Z (i, j)=x (i, j)+b (j)+r (i, j) 1≤i≤L, 1≤j≤N (10)
CFPN elimination comprises the concrete steps that, to all sample image Y1,Y2,…,YKThe new image of K width is obtained after carrying out RFPN correction,
It is denoted as Z1,Z2,…,ZK, ZkThe column mean vector of (1≤k≤K) is defined as Vk, then total column mean vector V of all K width new images
It can be expressed as formula (11):
Total column mean vector V is averaged by multiple repairing weld to be calculated, and eliminates the influence of random noise, therefore can be based on
V estimates the corresponding CFPN gray value b (j) (1≤j≤N) of jth column pixel,
In order to construct the smooth ideal column mean vector of a curve approximation first based on vector V estimation CFPN gray value b (j)
VC, then vector V subtracts vector VCObtained difference is the CFPN vector estimated；
Vector VCIt can be constructed in such a way that adjacent elements several in vector V are averaged, that is, use arest neighbors mean filter
(Nearest Neighbor Averaging Filter, NNAF) is constructed, the NNAF with parameter, as shown in formula (12),
To construct vector VC:
Wherein, W is variable, VWIt is the column mean vector constructed using the NNAF of design, and VW(j) be jth column pixel average ash
Angle value；
Based on VWThe CFPN vector of estimation is defined as BW, and the average gray value of jth column pixel is defined as BW(j), then vector BWIt can
To be calculated by formula (13):
BW=floor (V-VW)(13)
Vector BWThe average value of middle all elements is defined as BM,W, and BM,WAbsolute value be defined as BMA,W, such as formula (14) and formula (15)
It is shown:
BMA,W=BM,W(15)
When N is sufficiently large, element in CFPN vector obeys the Gaussian Profile that mean value is zero, therefore, the CFPN estimated to
The average gray value of all elements should be zero or be approximately zero in amount, then corresponds to BMA,WIt is worth the smallest BWVector be to CFPN to
The best estimate of amount, by best estimate vector BWIt is denoted as BS, therefore formula 16 can be obtained:
B (i)=BS(16)。
2. TDI-CMOS image sensor FPN bearing calibration as described in claim 1, characterized in that capturing sample image number
According to comprising the concrete steps that, it is assumed that the resolution ratio of TDI-CMOS image sensor is 8-bit, the acquisition K width under identical test environment
Average gray value is about 127, i.e., half saturated uniform light image is denoted as Y1,Y2,…,YK, this K width image is used as estimation
The sample data of FPN gray value, for appointing piece image Yk, formula (1) is expressed as formula (4):
Yk(i, j)=Xk(i,j)-a(i)+b(j)+rk(i,j)1≤k≤K(4)
Yk(i, j) is X as k (plain i, (ji), j) a (figure i) is as (figure i, as j) acquiring in 1 process k, K chooses half to b (j sample) this rk
The foundation for being saturated gray value is " EMVAStandard 1288 " standard.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201511034506.5A CN105681693B (en) | 2015-12-31 | 2015-12-31 | TDI-CMOS image sensor FPN bearing calibration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201511034506.5A CN105681693B (en) | 2015-12-31 | 2015-12-31 | TDI-CMOS image sensor FPN bearing calibration |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105681693A CN105681693A (en) | 2016-06-15 |
| CN105681693B true CN105681693B (en) | 2019-05-10 |
Family
ID=56298656
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201511034506.5A Expired - Fee Related CN105681693B (en) | 2015-12-31 | 2015-12-31 | TDI-CMOS image sensor FPN bearing calibration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105681693B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107730454B (en) * | 2016-08-10 | 2021-02-05 | 苏州中科天启遥感科技有限公司 | Linear-tracking-based method for removing horizontal stripe noise in linear array image |
| CN106993184B (en) * | 2017-03-29 | 2021-06-22 | 国芯微(重庆)科技有限公司 | Test method for CMOS sensor |
| CN106973287B (en) * | 2017-03-29 | 2021-05-11 | 嘉准传感科技（湖南）有限公司 | Test system for CMOS sensor |
| CN112565640B (en) * | 2020-10-30 | 2022-09-27 | 北京时代民芯科技有限公司 | Method and system for eliminating column noise of CMOS image sensor in nuclear environment |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102595060A (en) * | 2012-03-15 | 2012-07-18 | 天津大学 | Analog accumulator capable of implementing time delay integration (TDI) function inside complementary metal-oxide semiconductor (CMOS) image sensor |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7391925B2 (en) * | 2003-12-04 | 2008-06-24 | Lockheed Martin Missiles & Fire Control | System and method for estimating noise using measurement based parametric fitting non-uniformity correction |
-
2015
- 2015-12-31 CN CN201511034506.5A patent/CN105681693B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102595060A (en) * | 2012-03-15 | 2012-07-18 | 天津大学 | Analog accumulator capable of implementing time delay integration (TDI) function inside complementary metal-oxide semiconductor (CMOS) image sensor |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105681693A (en) | 2016-06-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105681693B (en) | TDI-CMOS image sensor FPN bearing calibration | |
| CN108981611B (en) | Digital projection grating image fitting correction method based on distortion global correction | |
| JP2016510108A5 (en) | ||
| US20160205376A1 (en) | Information processing apparatus, control method for the same and storage medium | |
| CN108871588B (en) | Multi-dimensional joint non-uniformity correction method for infrared imaging system | |
| JP2013038504A (en) | Imaging device, image processing method and program | |
| WO2016017450A1 (en) | Image-processing device, image processing method, program, and image sensor | |
| JP6161276B2 (en) | Measuring apparatus, measuring method, and program | |
| CN109389630B (en) | Method and device for determining and registering feature point set of visible light image and infrared image | |
| Qian et al. | Space low-pass and temporal high-pass nonuniformity correction algorithm | |
| KR20090061927A (en) | Defect and noise removal method | |
| TWI479454B (en) | Method and apparatus for correcting for vignetting in an imaging system | |
| US20170210403A1 (en) | Method to obtain data concerning the upper profile of an element of a railway track or switch | |
| FarshbafDoustar et al. | A locally-adaptive approach for image gamma correction | |
| Flesia et al. | Sub-pixel straight lines detection for measuring through machine vision | |
| CN113432723B (en) | Image processing method, system and computer system for weakening stray radiation | |
| CN114630009A (en) | Noise and tailing removing method for pulse sequence type image sensor | |
| Gong et al. | User-aided single image shadow removal | |
| CN106506999B (en) | Moment matching-based TDI-CMOS (time delay integration-complementary metal oxide semiconductor) image sensor FPN (field-programmable gate array) correction method | |
| US8576310B2 (en) | Image processing apparatus, camera module, and image processing method | |
| Jara Chavez et al. | Acceleration algorithm for constant-statistics method applied to the nonuniformity correction of infrared sequences | |
| CN103985089B (en) | With reference to weight edge analysis and the image streak correction method of frame inner iteration | |
| JPWO2020217283A5 (en) | Object detection device, object detection system, object detection method and program | |
| JP2019075658A5 (en) | ||
| CN104394334B (en) | Intelligent suppression method for on-orbit noise of image sensor |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190510 Termination date: 20201231 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |