CN105430385B - The method and device that a kind of dead pixels of image sensor is surveyed and corrected - Google Patents
The method and device that a kind of dead pixels of image sensor is surveyed and corrected Download PDFInfo
- Publication number
- CN105430385B CN105430385B CN201510929067.8A CN201510929067A CN105430385B CN 105430385 B CN105430385 B CN 105430385B CN 201510929067 A CN201510929067 A CN 201510929067A CN 105430385 B CN105430385 B CN 105430385B
- Authority
- CN
- China
- Prior art keywords
- point
- bad point
- bad
- data
- gain
- 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.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N17/00—Diagnosis, testing or measuring for television systems or their details
- H04N17/002—Diagnosis, testing or measuring for television systems or their details for television cameras
-
- 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
- H04N25/68—Noise processing, e.g. detecting, correcting, reducing or removing noise applied to defects
- H04N25/683—Noise processing, e.g. detecting, correcting, reducing or removing noise applied to defects by defect estimation performed on the scene signal, e.g. real time or on the fly detection
Abstract
A kind of method and device surveyed the invention discloses dead pixels of image sensor and corrected, device includes:Vision sensor data input block, gain compensation unit, bad point judging unit, bad point intensity control unit, bad point correction is gone unit and to remove bad point data outputting unit.The data that the various arrangements of four filters of the present invention using 2x2 as cycle are produced carry out gain calculating and gain compensation as input data, the data window to setting up MxN centered on each point of input data;Bad point situation for each all directions put of input data is judged, and obtains bad point information;Propose that the model of variance/brightness determines bad point intensity according to the texture information that each is put;Using bad point information and going bad point intensity to be corrected bad point, the data after output calibration.The bad point detection of the vision sensor data form of the various arrangement modes of four filters of the satisfaction with 2x2 as cycle and correction, can automatically and efficiently carry out bad point detection and correction.
Description
Technical field
The present invention relates to field of video image processing, more particularly to a kind of method that dead pixels of image sensor is surveyed and corrected
And device.
Background technology
Due to semiconductor fabrication process and the difference of raw material, can all there are some bad points, these bad points in imageing sensor
Mainly divide two types:A kind of is the bright always or dark fixed bad point of pixel;One kind is the time-varying side of pixel
Difference is bigger than the pixel of surrounding.
The bad point detection of conventional imageing sensor and the method for correction have two major classes at present:One class is by offline mode
Count the bad point position of imageing sensor, this method bad point position in view of imageing sensor and intensity can be with
The change of temperature and time and change;Another kind of is to carry out bad point detection and correction to imageing sensor in real time.
The imageing sensor of the filter arrangement mode based on 2x2 has four kinds of main types in the market, and such as Fig. 1 a~
Shown in 1d:Bayer format (such as Fig. 1 a), RGBIR forms (such as Fig. 1 b), RGBW forms (such as Fig. 1 c) and CyYeMgGr forms are (such as
Fig. 1 d), simply a kind of arrangement mode therein shown in Fig. 1, other modes be the position of R and B, the position of G and IR or W, Cy with
The position of Gr and the position of Ye and Mg can exchange.It is not a kind of at present for 2x2 be in the minimum sampling period four filter
Bad point detection and the method for correction that the various different arrangement modes of mirror are all suitable for, the present invention disclosure satisfy that with 2x2 as cycle
The bad point detection of four vision sensor data forms of the various arrangement modes of filter and correction.
The content of the invention
The technical problems to be solved by the invention are to provide the method and dress of a kind of dead pixels of image sensor survey and correction
Put, disclosure satisfy that the bad point detection of the vision sensor data form of the various arrangement modes of four filters with 2x2 as cycle
And correction, while extracting control image removes the model of bad point intensity, so as to it is automatic and it is efficient carry out bad point detection and
Correction.
To realize above-mentioned technique effect, the device surveyed the invention discloses a kind of dead pixels of image sensor and corrected, bag
Include:
One vision sensor data input block, for input picture sensing data;
One gain compensation unit, is connected with described image sensor data input cell, for being sensed with described image
The data of each passage in MxN data windows centered on each point of device data carry out gain calculating and gain compensation;
One bad point judging unit, is connected with the gain compensation unit, for judging through gain calculating and gain compensation
Whether each point of vision sensor data meets bad point condition in all directions of the MxN data windows, and will determine that knot
Fruit is used as bad point information output;
One removes bad point intensity control unit, is connected with the gain compensation unit, for calculating described image sensor number
According to each point centered on MxN data windows in current point with the point of passage variance and mean flow rate, using the variance
Texture strength with the mean flow rate calculates the current point, current point is determined according to the texture strength being calculated
Go bad point intensity;
One bad point corrects unit, and the gain compensation unit, the bad point judging unit and described goes bad point intensity control
Unit processed connection, for according to the bad point information and it is described go bad point Strength co-mputation to obtain bad point corrected value, and using described
The correction of bad point corrected value is described through gain calculating and the vision sensor data of gain compensation;And
One removes bad point data outputting unit, is connected with bad point correction unit, for the image sensing after output calibration
Device data.
Described image sensor bad point detection and the device of correction further improve and are, described image sensor data
It is data that the various arrangements of four filters with 2x2 as cycle are produced, the form of described image sensor data is Bayer lattice
Any one in formula, RGBIR forms, RGBW forms and CyYeMgGr forms.
Described image sensor bad point detection and the device of correction further improve and are, the gain compensation unit bag
Include:
One gain calculation module, is connected with described image sensor data input cell, for by with MxN data windows
The ratio of luminance mean value of luminance mean value and first passage of each passage the gain of compensation is needed as each passage, carry out
Gain is calculated;And
One gain compensation block, is connected with the gain calculation module, right for being multiplied by by the data to each passage
Answering passage needs the gain of compensation, carries out gain compensation.
Described image sensor bad point detection and the device of correction further improve and are, the bad point judging unit bag
Include:
One bad point horizontal direction judge module, is connected with the gain compensation unit, is calculated through gain and increased for judging
Whether each point of the vision sensor data of benefit compensation meets bad point condition in the horizontal direction of the MxN data windows;
One bad point vertical direction judge module, is connected with the gain compensation unit, is calculated through gain and increased for judging
Whether each point of the vision sensor data of benefit compensation meets bad point condition in the vertical direction of the MxN data windows;
One 45 degree of bad point walking direction module, is connected with the gain compensation unit, is calculated through gain and increased for judging
Whether each point of the vision sensor data of benefit compensation meets bad point condition on 45 degree of directions of the MxN data windows;
One 135 degree of bad point walking direction module, is connected with the gain compensation unit, is calculated through gain and increased for judging
Whether each point of the vision sensor data of benefit compensation meets bad point condition on 135 degree of directions of the MxN data windows;With
And
One bad point entirety judge module, mould is judged with the bad point horizontal direction judge module, the bad point vertical direction
The correction unit connection of block, 45 degree of walking direction modules of the bad point, 135 degree of walking direction modules of the bad point and the bad point,
For judging whether each point through the vision sensor data of gain calculating and gain compensation is met in the MxN data simultaneously
Four bad point conditions on the horizontal direction of window, vertical direction, 45 degree of directions and 135 degree of directions, will simultaneously meet four bad points
The situation of condition is judged as bad point, and the situation that four bad point conditions are met when will be different is judged as normal point, will on bad point and
The judged result of normal point corrects unit as bad point information conveyance to the bad point.
Described image sensor bad point detection and the device of correction further improve and are, described to go bad point strength control
Unit includes:
One variance computing module, is connected with the gain compensation unit, for calculating the every of described image sensor data
In the MxN data windows of individual point with current point with the point of passage variance;
One mean flow rate computing module, is connected with the gain compensation unit, for calculating described image sensor data
Each point MxN data windows in current point with the point of passage mean flow rate;And
One goes bad point Strength co-mputation module, with the variance computing module, the mean flow rate computing module and described bad
Point calibration unit is connected, the texture strength for being calculated current point according to the variance and the mean flow rate, and according to
Formula (a) calculates texture strength and removes bad point intensity W between texture strength lower threshold value and texture strength upper threshold value;
W=(SL-thl)/(thh-thl) (a)
Wherein, SL represents texture strength, and thl represents texture strength lower threshold value, and thh represents texture strength upper threshold value;
When the texture strength is less than texture strength lower threshold value, bad point intensity is gone for same with current point in MxN data windows
The mean flow rate of the point of passage;
When the texture strength is more than texture strength upper threshold value, the brightness value for current point is more than in MxN data windows
With current point with the brightness maxima of the point of passage situation, go bad point intensity in MxN data windows with current point with passage
Brightness maxima;For current point brightness value less than in MxN data windows with current point with passage brightness minimum value in the case of,
Go bad point intensity in MxN data windows with current point with passage brightness minimum value;For current point brightness value in MxN data
The situation between brightness maxima and brightness minimum value in window with current point with passage, goes the brightness that bad point intensity is current point
Value.
Described image sensor bad point detection and the device of correction further improve and are, the bad point corrects unit bag
Include:
One maximum value calculation module, and the gain compensation unit, the bad point judging unit and described goes bad point intensity
Control unit connect, for calculate described image sensor data each point MxN data windows in current point with passage
The brightness maxima of point;
One minimum value computing module, and the gain compensation unit, the bad point judging unit and described goes bad point intensity
Control unit connect, for calculate described image sensor data each point MxN data windows in current point with passage
The brightness minimum value of point;
One mean value computation module, and the gain compensation unit, the bad point judging unit and described goes bad point intensity control
Unit processed connection, in the MxN data windows for each point for calculating described image sensor data with current point with passage point
Mean flow rate;And
One bad point correct intensity control module, with the maximum value calculation module, the minimum value computing module, it is described
Value computing module and it is described go bad point data outputting unit connect, for according to the brightness maxima, the brightness minimum value,
The mean flow rate, bad point information and bad point Strength co-mputation is gone to obtain bad point corrected value, and using bad point corrected value correction
It is described through gain calculating and the vision sensor data of gain compensation.
A kind of method surveyed the invention also discloses dead pixels of image sensor and corrected, including:
Input picture sensing data;
Data to each passage in the MxN data windows centered on each point of described image sensor data are carried out
Gain calculating and gain compensation;
Judge through the vision sensor data of gain calculating and gain compensation each point the MxN data windows each
Whether bad point condition is met on direction, and will determine that result as bad point information output;
With current point with the point of passage in MxN data windows centered on each point of calculating described image sensor data
Variance and mean flow rate, the texture strength of the current point is calculated using the variance and the mean flow rate, according to calculating
To the texture strength determine that current point goes bad point intensity;
According to the bad point information and it is described go bad point Strength co-mputation to obtain bad point corrected value, and using bad point correction
Value correction is described through gain calculating and the vision sensor data of gain compensation;And
Vision sensor data after output calibration.
Described image sensor bad point detection and the method for correction are further improved and are, described image sensor data
It is data that the various arrangements of four filters with 2x2 as cycle are produced, the form of described image sensor data is Bayer lattice
Any one in formula, RGBIR forms, RGBW forms and CyYeMgGr forms.
Described image sensor bad point detection and the method for correction are further improved and are, to described image sensor
The data of each passage in MxN data windows centered on each point of data carry out gain calculating and gain compensation, further
Including:
Using the ratio of the luminance mean value of each passage in the MxN data windows and the luminance mean value of first passage as each
Individual passage needs the gain of compensation, carries out gain calculating;And
Data to each passage are multiplied by the gain that respective channel needs to compensate, and carry out gain compensation.
It is to judge to be calculated through gain and increased that described image sensor bad point detection and the method for correction are further improved
Whether each point of the vision sensor data of benefit compensation meets bad point condition in all directions of the MxN data windows, and
Will determine that result, as bad point information output, is further included:
Judge through the vision sensor data of gain calculating and gain compensation each point the MxN data windows level
Whether bad point condition is met on direction;
Judge each point through the vision sensor data of gain calculating and gain compensation in the vertical of the MxN data windows
Whether bad point condition is met on direction;
Judge each point through the vision sensor data of gain calculating and gain compensation at 45 degree of the MxN data windows
Whether bad point condition is met on direction;
Judge each point through the vision sensor data of gain calculating and gain compensation the 135 of the MxN data windows
Whether degree meets bad point condition on direction;And
Judge whether meet in the MxN simultaneously through each point of the vision sensor data of gain calculating and gain compensation
Four bad point conditions on the horizontal direction of data window, vertical direction, 45 degree of directions and 135 degree of directions, will simultaneously meet four
The situation of bad point condition is judged as bad point, and the situation that four bad point conditions are met when will be different is judged as normal point, will be on bad
The judged result of point and normal point is used as bad point information output.
Described image sensor bad point detection and the method for correction are further improved and are, calculate described image sensor
Data each point centered on MxN data windows in current point with the point of passage variance and mean flow rate, using the side
Difference and the mean flow rate calculate the texture strength of the current point, work as according to the texture strength being calculated is determined
Preceding point goes bad point intensity, further includes:
With current point with the point of passage in MxN data windows centered on each point of calculating described image sensor data
Variance;
With current point with the point of passage in MxN data windows centered on each point of calculating described image sensor data
Mean flow rate;And
The texture strength of current point is calculated according to the variance and the mean flow rate, and is calculated according to formula (a)
Texture strength removes bad point intensity W between texture strength lower threshold value and texture strength upper threshold value;
W=(SL-thl)/(thh-thl) (a)
Wherein, SL represents texture strength, and thl represents texture strength lower threshold value, and thh represents texture strength upper threshold value;
When the texture strength is less than texture strength lower threshold value, bad point intensity is gone for same with current point in MxN data windows
The mean flow rate of the point of passage;
When the texture strength is more than texture strength upper threshold value, the brightness value for current point is more than in MxN data windows
With current point with the brightness maxima of the point of passage situation, go bad point intensity in MxN data windows with current point with passage
Brightness maxima;For current point brightness value less than in MxN data windows with current point with passage brightness minimum value in the case of,
Go bad point intensity in MxN data windows with current point with passage brightness minimum value;For current point brightness value in MxN data
The situation between brightness maxima and brightness minimum value in window with current point with passage, goes the brightness that bad point intensity is current point
Value.
Described image sensor bad point detection and the method for correction are further improved and are, according to the bad point information and
It is described to go bad point Strength co-mputation to obtain bad point corrected value, and correct described through gain calculating and gain using the bad point corrected value
The vision sensor data of compensation, further includes:
With current point with the point of passage in MxN data windows centered on each point of calculating described image sensor data
Brightness maxima;
With current point with the point of passage in MxN data windows centered on each point of calculating described image sensor data
Brightness minimum value;
With current point with the point of passage in MxN data windows centered on each point of calculating described image sensor data
Mean flow rate;And
According to the brightness maxima, the brightness minimum value, the mean flow rate, bad point information and remove bad point intensitometer
Calculation obtains bad point corrected value, and described through gain calculating and the imageing sensor of gain compensation using bad point corrected value correction
Data.
The present invention makes it have following beneficial effect as a result of above technical scheme:
The present invention proposes that control image goes bad point on the basis of bad point detection and correction is carried out to imageing sensor in real time
The model of intensity is so as to self adaptation according to scene information determines the intensity of bad point.Disclosure satisfy that with 2x2 as cycle
The dead pixel points of images of data form of various arrangement modes of four filters eliminate function, model according to variance/brightness is controlled
The imaged intensity for going bad point, can in real time and efficiently carry out bad point elimination, meet change and the demand of various different scenes.
Brief description of the drawings
Fig. 1 a~1d is four kinds of forms of the imageing sensor of the existing filter arrangement mode for being based on 2 × 2.
Fig. 2 is the structural representation of the device of a kind of dead pixels of image sensor survey of the invention and correction.
Fig. 3 is the structural representation of the gain compensation unit of the device of a kind of dead pixels of image sensor survey of the invention and correction
Figure.
Fig. 4 is the structural representation of the bad point judging unit of the device of a kind of dead pixels of image sensor survey of the invention and correction
Figure.
Fig. 5 be a kind of dead pixels of image sensor of the invention survey and correction device bad point judging unit in current point 5
The information schematic diagram of × 5 data windows.
Fig. 6 is the knot for removing bad point intensity control unit of the device of a kind of dead pixels of image sensor survey of the invention and correction
Structure schematic diagram.
Fig. 7 is the structural representation of the bad point correction unit of the device of a kind of dead pixels of image sensor survey of the invention and correction
Figure.
Fig. 8 is the flow chart of the method for a kind of dead pixels of image sensor survey of the invention and correction.
Fig. 9 is the flow chart of the preferred embodiment of the method for a kind of dead pixels of image sensor survey of the invention and correction.
Specific embodiment
Below in conjunction with the accompanying drawings and specific embodiment the present invention is further detailed explanation.
The present invention provides a kind of imageing sensor of the various arrangement modes of four filters that disclosure satisfy that with 2x2 as cycle
The bad point detection of data form and the method and apparatus of correction, while the present invention controls image according to the model of variance/brightness
Bad point intensity is gone, be capable of self adaptation determines the intensity of bad point according to scene information.
Referring initially to shown in Fig. 2, dead pixels of image sensor of the present invention is surveyed and the device of correction is mainly passed including an image
Sensor data input cell 11, a gain compensation unit 12, a bad point judging unit 13, go bad point intensity control unit 14,
One bad point correction unit 15 and removes bad point data outputting unit 16.
Vision sensor data input block 11 is an imageing sensor, for input picture sensing data, wherein,
The data that the various arrangements of four filters with 2x2 as cycle are produced, described image sensor data form be Bayer format,
Any one in RGBIR forms, RGBW forms and the type of CyYeMgGr forms four, vision sensor data input block
The vision sensor data that the 11 various arrangements for being also responsible for four filters with 2x2 as cycle to follow-up unit transportation are produced.
Gain compensation unit 12 is connected with vision sensor data input block 11, for the imageing sensor to be input into
The data of each passage in the MxN data windows set up centered on each point of data carry out gain calculating and gain compensation, will
The mean flow rate of each passage is adjusted to approximately equalised degree.
The vision sensor data that the various arrangements of four filters for input with 2x2 as cycle are produced, gain compensation
Unit 12 is using four channel gain compensation.In order to the information of all passages can be used in bad point judging unit 13, it is necessary to will
The mean flow rate of four passages is adjusted to approximately equalised degree, so needing to carry out simply the vision sensor data being input into
Gain calculating and gain compensation.
In order to realize conveniently, 5x5 data windows are set up centered on each point of image taking sensor data, and by the 5x5 numbers
The gain of compensation is needed as each passage according to the ratio of four averages of passage in window and the average of first passage, in general
The value of the gain that data window is compensated the need for being calculated more greatly is also more accurate.
Refering to shown in Fig. 3, gain compensation unit 12 further includes a gain calculation module 121 and a gain compensation block
122.Wherein, gain calculation module 121 is connected with vision sensor data input block 11, for receiving vision sensor data
The vision sensor data that the conveying of input block 11 comes, and by with four luminance mean values of passage in above-mentioned 5x5 data windows
Ratio with the luminance mean value of first passage needs the gain of compensation as each passage, carries out gain calculating.Gain compensation mould
Block 122 is connected with gain calculation module 121, and being directly multiplied by respective channel for the input data to imageing sensor needs compensation
Gain, carry out gain compensation.
Bad point judging unit 13 is connected with the gain compensation block 122 of gain compensation unit 12, is mainly used in judging through increasing
Benefit is calculated with whether each point of the vision sensor data of gain compensation meets bad point bar in all directions of 5x5 data windows
Part, and will determine that result as bad point information output.
Refering to shown in Fig. 4 and Fig. 5, bad point judging unit 13 further includes a bad point horizontal direction judge module 131,
Bad point vertical direction judge module 132,45 degree of walking direction modules 133 of a bad point, 135 degree of walking direction modules 134 of a bad point
And one bad point entirety judge module 135.For setting up as shown in Figure 5 centered on each point of imageing sensor input data
5x5 data windows.
Bad point horizontal direction judge module 131 is connected with the gain compensation block 122 of gain compensation unit 12, for judging
Whether meet the level in above-mentioned 5x5 data windows through the current point D33 of gain calculating and the vision sensor data of gain compensation
Bad point condition on direction.Wherein, bad point horizontal direction determination methods are:If meeting the bar of formula (1) and formula (2) simultaneously
Part, then bad point condition is met in this horizontal direction.
Bad point vertical direction judge module 132 is connected with the gain compensation block 122 of gain compensation unit 12, for judging
Whether meet in the vertical of above-mentioned 5x5 data windows through the current point D33 of gain calculating and the vision sensor data of gain compensation
Bad point condition on direction.Wherein, bad point vertical direction determination methods are:If meeting the bar of formula (3) and formula (4) simultaneously
Part, then bad point condition is met in this vertical direction.
45 degree of walking direction modules 133 of bad point are connected with the gain compensation block 122 of gain compensation unit 12, for judging
Whether 45 degree in above-mentioned 5x5 data windows are met through the current point D33 of gain calculating and the vision sensor data of gain compensation
Bad point condition on direction.Wherein, 45 degree of direction determination process of bad point are:If meeting the bar of formula (5) and formula (6) simultaneously
Part, then this point meets bad point condition on 45 degree of directions.
abs(D33-D15)>max(abs(D32-D14),abs(D43-D25),abs(D42-D24)) (5)
abs(D33-D51)>max(abs(D34-D52),abs(D23-D41),abs(D24-D42)) (6)
135 degree of walking direction modules 134 of bad point are connected with the gain compensation block 122 of gain compensation unit 12, for sentencing
Whether the current point D33 of the vision sensor data of cracked ends gain calculating and gain compensation is met the 135 of above-mentioned 5x5 data windows
Bad point condition on degree direction.Wherein, 135 degree of direction determination process of bad point are:If meeting formula (7) and formula (8) simultaneously
Condition, then this point meets bad point condition on 135 degree of directions.
abs(D33-D11)>max(abs(D34-D12),abs(D43-D21),abs(D44-D22)) (7)
abs(D33-D55)>max(abs(D32-D54),abs(D23-D45),abs(D22-D44)) (8)
Bad point horizontal direction judge module 131 above, bad point vertical direction judge module 132,45 degree of directions of bad point are sentenced
135 degree of walking direction modules 134 of disconnected module 133 and bad point are the bad point judge module on four direction arranged side by side, it is therefore intended that
Judge whether this point of the vision sensor data through gain calculating and gain compensation is met in all directions of 5x5 data windows
Bad point condition, and will determine that result is delivered to next module jointly:Bad point entirety judge module 135.
Bad point entirety judge module 135 and bad point horizontal direction judge module 131, bad point vertical direction judge module 132,
45 degree of walking direction modules 133 of bad point, 135 degree of walking direction modules 134 of bad point and bad point correction unit 15 are connected, for judging
Whether each point through the vision sensor data of gain calculating and gain compensation is met in the level side of 5x5 data windows simultaneously
To four bad point conditions on, vertical direction, 45 degree of directions and 135 degree of directions, four situations of bad point condition will be simultaneously met
It is judged as bad point, other situations that four bad point conditions are met when will be different are judged as normal point, i.e., be not bad point, will close
In the judged result of bad point and normal point unit 15 is corrected as bad point information conveyance to bad point.
Go bad point intensity control unit 14 to be connected with the gain compensation block 122 of gain compensation unit 12, be mainly used in meter
In calculating 5x5 data windows, in addition to current point and current point with the point of passage variance and mean flow rate, using being calculated
Variance and mean flow rate calculate the texture strength of current point current scene again, then according to the texture of the current point being calculated
Intensity determines that current point goes bad point intensity.
The present invention determines the intensity of bad point in order to self adaptation according to scene information, proposes that control image goes bad
The model of point intensity:Variance/brightness (sigma/L).The model of variance/brightness can distinguish current region be texture region also
It is flat site, if the intensity that current region is the correction of flat site bad point can be very strong, if texture region bad point school
Positive intensity needs to be changed according to the intensity of texture, and texture strength is stronger, and bad point correction intensity is weaker.
Further include that a variance computing module 141, is average bright refering to bad point intensity control unit 14 shown in Fig. 6, is removed
Degree computing module 142 and goes bad point Strength co-mputation module 143.
Wherein, variance computing module 141 is connected with the gain compensation block 122 of gain compensation unit 12, for calculating 5x5
In data window with current point D33 with the point of passage variance.Because current point D33 is likely to bad point, removes work as here
The influence of preceding point D33, be with the point of passage with current point D33 in 5x5 data windows:D11, D13, D15, D31, D35, D51, D53,
D55, variance calculates such as formula (9):
Sigma=[std (D11, D13, D15, D31, D35, D51, D53, D55)]2 (9)
Mean flow rate computing module 142 is connected with the gain compensation block 122 of gain compensation unit 12, for calculating 5x5
In data window with current point D33 with the point of passage mean flow rate.The influence of current point is removed, it is flat with passage in 5x5 data windows
Equal brightness calculation such as formula (10):
Lmean=mean (D11, D13, D15, D31, D35, D51, D53, D55) (10)
Bad point Strength co-mputation module 143 is gone to be corrected with variance computing module 141, mean flow rate computing module 142 and bad point
Unit 15 is connected, and the variance and average brightness calculation module 142 for being calculated according to variance computing module 141 are calculated
Each point texture strength (SL).With the variance of passage in the current point 5x5 windows that texture strength (SL) is obtained equal to formula (9)
In the current point 5x5 windows obtained with formula (10) with passage mean flow rate ratio, calculate such as formula (11):
SL=sigma/Lmean (11)
Going bad point Strength co-mputation module 143 can determine bad point intensity according to the texture strength of current scene, if line
Reason intensity is less than texture strength lower threshold value (thl), and texture strength is very weak (i.e. flat site), goes bad point intensity to use very strong
Intensity, go bad point intensity use current point 5x5 data windows in current point with the point of passage mean flow rate.If texture strength
More than texture strength upper threshold value (thh), texture strength is very strong, and bad point correction needs very weak intensity, for the brightness of current point
Value more than in 5x5 data windows with current point with the brightness maxima of the point of passage situation, it is just 5x5 data windows to go bad point intensity
In with current point with the point of passage brightness maxima;Brightness value for current point is same with current point less than in 5x5 data windows
The situation of the brightness minimum value of the point of passage, go bad point intensity just in 5x5 data windows with current point with the point of passage brightness
Minimum value;For current point brightness value in 5x5 data windows with current point with the brightness maxima of the point of passage and brightness most
Value between small value avoids the need for changing, it is not necessary to carry out the correction of bad point.
If texture strength (SL) is between texture strength lower threshold value (thl) and texture strength upper threshold value (thh), then line
Reason intensity (SL) can be calculated bad point intensity according to texture strength lower threshold value (thl) and texture strength upper threshold value (thh)
(W) it is, specific to calculate such as formula (a):
W=(SL-thl)/(thh-thl) (a)
Bad point correction unit 15 respectively with gain compensation unit 12, bad point judging unit 13 and remove bad point intensity control unit
14 output connection, be mainly used according to bad point judging unit 13 conveying come bad point information with remove bad point intensity control unit 14
What conveying came goes bad point Strength co-mputation to obtain bad point corrected value, and is mended through gain calculating and gain using bad point corrected value correction
The vision sensor data repaid.
Refering to shown in Fig. 7, bad point correction unit 15 further includes that a maximum value calculation module 151, a minimum value are calculated
Module 152, a mean value computation module 153 and a bad point correction intensity control module 154.
Wherein, maximum value calculation module 151 and gain compensation unit 12, bad point judging unit 13 and bad point strength control is gone
The output connection of unit 14, for calculate in 5x5 data windows with current point with the point of passage brightness maxima.Removal is needed to work as
The influence of preceding point, such as formula (12) is calculated in 5x5 data windows with current point with the brightness maxima of the point of passage:
Lmax=max (D11, D13, D15, D31, D35, D51, D53, D55) (12)
Minimum value computing module 152 equally with gain compensation unit 12, bad point judging unit 13 and go bad point strength control
The output connection of unit 14, for calculate in 5x5 data windows with current point with the point of passage brightness minimum value.Removal is needed to work as
The influence of preceding point, such as formula (13) is calculated in 5x5 data windows with current point with the brightness minimum value of the point of passage:
Lmin=min (D11, D13, D15, D31, D35, D51, D53, D55) (13)
Mean value computation module 153 equally with gain compensation unit 12, bad point judging unit 13 and remove bad point strength control list
The output connection of unit 14, for calculate in 5x5 data windows with current point with the point of passage mean flow rate.Need to remove current point
Influence, calculated such as formula (14) with the mean flow rate of the point of passage with current point in 5x5 data windows:
Lmean=mean (D11, D13, D15, D31, D35, D51, D53, D55) (14)
Bad point corrects intensity control module 154 and maximum value calculation module 151, minimum value computing module 152, mean value computation
Module 152 and bad point data outputting unit 16 is gone to connect, for according to brightness maxima, brightness minimum value, mean flow rate, bad point
Information and go bad point Strength co-mputation to obtain bad point corrected value, and calculate and increase through gain using bad point corrected value correction is described
The vision sensor data of benefit compensation.
For current point be judged as bad point simultaneously current point brightness value more than in 5x5 data windows with current point with passage
The brightness maxima of point, the calculating such as formula (15) of bad point corrected value (DPC):
DPC=Lmean* (1-W)+Lmax*W (15)
For current point be judged as bad point simultaneously current point brightness value less than in 5x5 data windows with current point with passage
The brightness minimum value of point, the calculating such as formula (16) of bad point corrected value (DPC):
DPC=Lmean* (1-W)+Lmin*W (16)
For current point (CP) be judged as bad point simultaneously current point brightness value in 5x5 data windows with the same passage of current point
Point brightness maxima and brightness minimum value between value, the calculating such as formula (17) of bad point corrected value (DPC):
DPC=Lmean* (1-W)+CP*W (17)
It is judged as it not being bad point for current point, is processed without carrying out bad point, directly exports the brightness value of current point.
Finally, bad point data outputting unit 16 is gone to connect with the bad point correction intensity control module 154 of bad point correction unit 15
Connect, for receiving the next vision sensor data through the correction of bad point corrected value of the bad point correction conveying of intensity control module 154 simultaneously
It is conveyed to subsequent video images processing unit.
The device of dead pixels of image sensor survey of the present invention and correction disclosure satisfy that each of four filters with 2x2 as cycle
The dead pixel points of images for planting the data form of arrangement mode eliminates function, and the present invention controls image always according to the model of variance/brightness
The intensity of bad point is gone, bad point elimination can in real time and be efficiently carried out, change and the demand of various different scenes is met.
Refering to shown in Fig. 8, dead pixels of image sensor of the present invention is surveyed and the method for correction mainly includes:
S001:Input picture sensing data;
S002:Data to each passage in the MxN data windows centered on each point of image taking sensor data are entered
Row gain calculating and gain compensation;
S003:Judge each point through the vision sensor data of gain calculating and gain compensation in each of MxN data windows
Whether bad point condition is met on individual direction, and will determine that result as bad point information output;
S004:Calculate vision sensor data each point centered on MxN data windows in current point with passage point
Variance and mean flow rate, it is strong according to the texture being calculated using variance and the texture strength of average brightness calculation current point
That spends decision current point goes bad point intensity;
S005:According to bad point information and bad point Strength co-mputation is gone to obtain bad point corrected value, and using the correction of bad point corrected value
Through gain calculating and the vision sensor data of gain compensation;And
S006:Vision sensor data after output calibration.
Wherein, the vision sensor data of input is the data of the various arrangements generation of four filters with 2x2 as cycle,
Vision sensor data form is any one in Bayer format, RGBIR forms, RGBW forms and CyYeMgGr forms.
Step S002:To the number of each passage in the MxN data windows centered on each point of image taking sensor data
According to gain calculating and gain compensation is carried out, further include:
It is logical as each using the ratio of the luminance mean value of each passage in MxN data windows and the luminance mean value of first passage
Road needs the gain of compensation, carries out gain calculating;And
Being multiplied by respective channel to the data of current point each passage needs the gain of compensation, carries out gain compensation.
Step S003:Judge each point through the vision sensor data of gain calculating and gain compensation in MxN data windows
All directions on whether meet bad point condition, and will determine that result, as bad point information output, is further included:
Judge through the vision sensor data of gain calculating and gain compensation each point MxN data windows horizontal direction
On whether meet bad point condition;
Judge through the vision sensor data of gain calculating and gain compensation each point MxN data windows vertical direction
On whether meet bad point condition;
Judge each point through the vision sensor data of gain calculating and gain compensation in 45 degree of directions of MxN data windows
On whether meet bad point condition;
Judge each point through the vision sensor data of gain calculating and gain compensation in 135 degree of sides of MxN data windows
Whether meet bad point condition upwards;And
Judge whether meet in MxN data simultaneously through each point of the vision sensor data of gain calculating and gain compensation
Four bad point conditions on the horizontal direction of window, vertical direction, 45 degree of directions and 135 degree of directions, will simultaneously meet four bad points
The situation of condition is judged as bad point, and the situation that four bad point conditions are met when will be different is judged as normal point, will on bad point and
The judged result of normal point is used as bad point information output.
Step S004:With the same passage of current point in MxN data windows centered on each point of calculating vision sensor data
Point variance and mean flow rate, using variance and the texture strength of average brightness calculation current point, according to the line being calculated
That manages intensity decision current point goes bad point intensity, further includes:
Calculate vision sensor data each point centered on MxN data windows in current point with the point of passage side
Difference;
With current point with the average of the point of passage in MxN data windows centered on each point of calculating vision sensor data
Brightness;And
The texture strength (SL) of current point is obtained according to variance and average brightness calculation, and texture is calculated according to formula (a)
Intensity (SL) goes bad point intensity (W) between texture strength lower threshold value (thl) and texture strength upper threshold value (thh);
W=(SL-thl)/(thh-thl) (a)
When texture strength is less than texture strength lower threshold value, go bad point intensity in MxN data windows with the same passage of current point
Point mean flow rate;
When texture strength is more than texture strength upper threshold value, it is more than in MxN data windows and works as the brightness value of current point
Preceding point with the brightness maxima of the point of passage situation, go bad point intensity in MxN data windows with current point with passage brightness
Maximum;For current point brightness value less than in MxN data windows with current point with passage brightness minimum value in the case of, go bad
Point intensity be MxN data windows in current point with passage brightness minimum value;For current point brightness value in MxN data windows
The situation between brightness maxima and brightness minimum value with current point with passage, removes the brightness value that bad point intensity is current point.
Step S005:According to bad point information and go bad point Strength co-mputation to obtain bad point corrected value, and utilize bad point corrected value
Correction is further included through gain calculating and the vision sensor data of gain compensation:
Calculate MxN data windows in current point with the point of passage brightness maxima;
Calculate MxN data windows in current point with the point of passage brightness minimum value;
Calculate MxN data windows in current point with the point of passage mean flow rate;And
According to brightness maxima, brightness minimum value, mean flow rate, bad point information and bad point Strength co-mputation is gone to obtain bad point school
On the occasion of, and using the correction of bad point corrected value through gain calculating and the vision sensor data of gain compensation.
The embodiment flow of the method that dead pixels of image sensor of the invention is surveyed and corrected with reference to Fig. 9 is done to be had
Body is introduced, and is implemented for convenience, MxN data windows 5 × 5 data for using the foundation centered on current point D33 as shown in Figure 5
Window.Dead pixels of image sensor of the present invention is surveyed and the method for correction is specifically included:
Step 201:Vision sensor data is input into.Conveying Bayer format or RGBIR forms or RGBW forms or
The data of CyYeMgGr forms, or the different arrangement modes of other four filters with 2x2 as cycle data.
Step 202:Four channel gains are calculated.Four averages of passage in 5x5 data windows centered on current point
Ratio with the average of first passage needs the gain of compensation as each passage.
Step 203, four channel gain compensation.Four data of passage are directly multiplied by with the correspondence that step 202 obtains to lead to
Road needs the gain of compensation.
Step 204:Bad point horizontal direction judges.If meeting the condition of formula (1) and (2) simultaneously, then current point exists
Bad point condition is met in the horizontal direction of 5x5 data windows.
Step 205:Bad point vertical direction judges.If meeting the condition of formula (3) and (4) simultaneously, then current point exists
Bad point condition is met in the vertical direction of 5x5 data windows.
Step 206:45 degree of walking directions of bad point.If meeting the condition of formula (5) and (6) simultaneously, then current point exists
Meet bad point condition on 45 degree of directions of 5x5 data windows.
abs(D33-D15)>max(abs(D32-D14),abs(D43-D25),abs(D42-D24)) (5)
abs(D33-D51)>max(abs(D34-D52),abs(D23-D41),abs(D24-D42)) (6)
Step 207:135 degree of walking directions of bad point.If meeting the condition of formula (7) and (8) simultaneously, then current point exists
Meet bad point condition on 135 degree of directions of 5x5 data windows.
abs(D33-D11)>max(abs(D34-D12),abs(D43-D21),abs(D44-D22)) (7)
abs(D33-D55)>max(abs(D32-D54),abs(D23-D45),abs(D22-D44)) (8)
Step 208:Bad point integrally judges.If current point 5x5 data windows level, it is vertical, 45 degree and 135 degree four
Direction all meets bad point condition, just thinks that current point is bad point, and other situations are not bad points.By the judged result on bad point
As bad point information output.
Step 209:Current 5x5 windows are calculated with passage variance.Calculate 5x5 data windows in current point D33 with passage point
Variance.Because current point D33 is likely to bad point, the influence of current point D33 is removed here, in 5x5 data windows with work as
Preceding point D33 is with the point of passage:D11, D13, D15, D31, D35, D51, D53, D55, variance are calculated as shown in formula (9).
Sigma=[std (D11, D13, D15, D31, D35, D51, D53, D55)]2 (9)
Step 210:Current 5x5 windows are with passage mean value computation.Calculate 5x5 data windows in current point D33 with passage point
Mean flow rate.The influence of current point is removed, is calculated as shown in formula (10) with passage mean flow rate in 5x5 data windows.
Lmean=mean (D11, D13, D15, D31, D35, D51, D53, D55) (10)
Step 211:Go bad point strength control.What the variance and formula (10) being calculated according to formula (9) were calculated
Mean flow rate is calculated the texture strength (SL) of current point D33.Texture strength (SL) is equal to the current point that formula (9) is obtained
In the current point 5x5 windows obtained with variance and the formula (10) of passage in 5x5 windows with passage mean flow rate ratio, calculate such as
Shown in formula (11).
SL=sigma/Lmean (11)
Going bad point Strength co-mputation module 143 can determine bad point intensity according to the texture strength of current scene, if line
Reason intensity is less than texture strength lower threshold value (thl), and texture strength is very weak (i.e. flat site), goes bad point intensity to use very strong
Intensity, go bad point intensity use current point 5x5 data windows in current point with the point of passage mean flow rate.If texture strength
More than texture strength upper threshold value (thh), texture strength is very strong, and bad point correction needs very weak intensity, for the brightness of current point
Value more than in 5x5 data windows with current point with the brightness maxima of the point of passage situation, it is just 5x5 data windows to go bad point intensity
In with current point with the point of passage brightness maxima;Brightness value for current point is same with current point less than in 5x5 data windows
The situation of the brightness minimum value of the point of passage, go bad point intensity just in 5x5 data windows with current point with the point of passage brightness
Minimum value;For current point brightness value in 5x5 data windows with current point with the brightness maxima of the point of passage and brightness most
Value between small value avoids the need for changing, and goes bad point intensity to think the brightness value of current point.
If texture strength (SL) is between texture strength lower threshold value (thl) and texture strength upper threshold value (thh), then line
Reason intensity (SL) can be calculated bad point intensity according to texture strength lower threshold value (thl) and texture strength upper threshold value (thh)
(W) it is, specific to calculate as shown in formula (a).
W=(SL-thl)/(thh-thl) (a)
Step 212:Current 5x5 windows are with passage maximum value calculation.With current point D33 with passage in calculating 5x5 data windows
The brightness maxima of point.Need to remove the influence of current point D33, in 5x5 data windows with current point D33 with the point of passage brightness
Shown in maximum value calculation such as formula (12).
Lmax=max (D11, D13, D15, D31, D35, D51, D53, D55) (12)
Step 213:Current 5x5 windows are calculated with passage minimum value.With current point D33 with passage in calculating 5x5 data windows
The brightness minimum value of point.Need to remove the influence of current point D33, in 5x5 data windows with current point D33 with the point of passage brightness
Minimum value is calculated as shown in formula (13).
Lmin=min (D11, D13, D15, D31, D35, D51, D53, D55) (13)
Step 214:Current 5x5 windows are with passage mean value computation.Calculate 5x5 data windows in current point D33 with passage point
Mean flow rate.Need to remove the influence of current point D33, in 5x5 data windows with current point D33 with the point of passage mean flow rate
Calculate as shown in formula (14).
Lmean=mean (D11, D13, D15, D31, D35, D51, D53, D55) (14)
Step 215:Bad point corrects strength control.What the bad point information and step 211 exported according to step 208 were exported goes bad
Intensity is put to determine the intensity of bad point correction.
It is judged as bad point for current point (CP) while the brightness value of current point is same with current point logical more than in 5x5 data windows
The situation of the brightness maxima of the point in road, shown in the calculating such as formula (15) of bad point corrected value (DPC);
DPC=Lmean* (1-W)+Lmax*W (15)
For current point be judged as bad point simultaneously current point brightness value less than in 5x5 data windows with current point with passage
The situation of the brightness minimum value of point, shown in the calculating such as formula (16) of bad point corrected value (DPC);
DPC=Lmean* (1-W)+Lmin*W (16)
For current point be judged as bad point simultaneously current point brightness value in 5x5 data windows with current point with passage point
Brightness maxima and brightness minimum value between value, shown in the calculating such as formula (17) of bad point corrected value (DPC);
DPC=Lmean* (1-W)+CP*W (17)
It is judged as it not being bad point for current point, is processed without carrying out bad point, directly exports the brightness value of current point.
The present invention is described in detail above in association with drawings and Examples, those skilled in the art can basis
Described above makes many variations example to the present invention.Thus, some of embodiment details should not constitute limitation of the invention,
The scope that to be defined using appended claims of the present invention is used as protection scope of the present invention.
Claims (12)
1. the device that a kind of dead pixels of image sensor is surveyed and corrected, it is characterised in that including:
One vision sensor data input block, for input picture sensing data;
One gain compensation unit, is connected with described image sensor data input cell, for described image sensor number
According to each point centered on MxN data windows in the data of each passage carry out gain calculating and gain compensation;
One bad point judging unit, is connected with the gain compensation unit, for judging the image through gain calculating and gain compensation
Whether each point of sensing data meets bad point condition in all directions of the MxN data windows, and will determine that result is made
It is bad point information output;
One removes bad point intensity control unit, is connected with the gain compensation unit, for calculating described image sensor data
Each point centered on MxN data windows in current point with the point of passage variance and mean flow rate, using the variance and institute
The texture strength that mean flow rate calculates current point is stated, goes bad point strong according to the texture strength decision current point being calculated
Degree;
One bad point corrects unit, and the gain compensation unit, the bad point judging unit and described removes bad point strength control list
Unit's connection, for according to the bad point information and it is described go bad point Strength co-mputation to obtain bad point corrected value, and utilize the bad point
Corrected value correction is described through gain calculating and the vision sensor data of gain compensation;And
One removes bad point data outputting unit, is connected with bad point correction unit, for the imageing sensor number after output calibration
According to.
2. the device that dead pixels of image sensor as claimed in claim 1 is surveyed and corrected, it is characterised in that:Described image is sensed
Device data are the data of the various arrangements generation of four filters with 2x2 as cycle, and the form of described image sensor data is
Any one in Bayer format, RGBIR forms, RGBW forms and CyYeMgGr forms.
3. the device that dead pixels of image sensor as claimed in claim 1 is surveyed and corrected, it is characterised in that the gain compensation
Unit includes:
One gain calculation module, is connected with described image sensor data input cell, for by with each in MxN data windows
The luminance mean value of individual passage needs the gain of compensation with the ratio of the luminance mean value of first passage as each passage, carries out gain
Calculate;And
One gain compensation block, is connected with the gain calculation module, logical for being multiplied by correspondence by the data to each passage
Road needs the gain of compensation, carries out gain compensation.
4. the device that dead pixels of image sensor as claimed in claim 1 is surveyed and corrected, it is characterised in that the bad point judges
Unit includes:
One bad point horizontal direction judge module, is connected with the gain compensation unit, for judging to be mended through gain calculating and gain
Whether each point for the vision sensor data repaid meets bad point condition in the horizontal direction of the MxN data windows;
One bad point vertical direction judge module, is connected with the gain compensation unit, for judging to be mended through gain calculating and gain
Whether each point for the vision sensor data repaid meets bad point condition in the vertical direction of the MxN data windows;
One 45 degree of bad point walking direction module, is connected with the gain compensation unit, for judging to be mended through gain calculating and gain
Whether each point for the vision sensor data repaid meets bad point condition on 45 degree of directions of the MxN data windows;
One 135 degree of bad point walking direction module, is connected with the gain compensation unit, for judging to be mended through gain calculating and gain
Whether each point for the vision sensor data repaid meets bad point condition on 135 degree of directions of the MxN data windows;And
One bad point entirety judge module, with the bad point horizontal direction judge module, the bad point vertical direction judge module, institute
The correction unit connection of 45 degree of walking direction modules of bad point, 135 degree of walking direction modules of the bad point and the bad point is stated, for sentencing
Whether each point of the vision sensor data of cracked ends gain calculating and gain compensation meets the water in the MxN data windows simultaneously
Square to four bad point conditions on, vertical direction, 45 degree of directions and 135 degree of directions, four bad point conditions will be simultaneously met
Situation is judged as bad point, and the situation that four bad point conditions are met when will be different is judged as normal point, will be on bad point and normal point
Judged result as bad point information conveyance to the bad point correct unit.
5. the device that dead pixels of image sensor as claimed in claim 1 is surveyed and corrected, it is characterised in that described to go bad point strong
Degree control unit includes:
One variance computing module, is connected with the gain compensation unit, each point for calculating described image sensor data
MxN data windows in current point with the point of passage variance;
One mean flow rate computing module, is connected with the gain compensation unit, for calculating the every of described image sensor data
In the MxN data windows of individual point with current point with the point of passage mean flow rate;And
One goes bad point Strength co-mputation module, with the variance computing module, the mean flow rate computing module and the bad point school
Positive unit connection, the texture strength for being calculated current point according to the variance and the mean flow rate, and according to formula
A () calculates texture strength and removes bad point intensity W between texture strength lower threshold value and texture strength upper threshold value;
W=(SL-thl)/(thh-thl) (a)
Wherein, SL represents texture strength, and thl represents texture strength lower threshold value, and thh represents texture strength upper threshold value;
When the texture strength is less than texture strength lower threshold value, go bad point intensity in MxN data windows with the same passage of current point
Point mean flow rate;
When the texture strength is more than texture strength upper threshold value, it is more than in MxN data windows and works as the brightness value of current point
Preceding point with the brightness maxima of the point of passage situation, go bad point intensity in MxN data windows with current point with passage brightness
Maximum;For current point brightness value less than in MxN data windows with current point with passage brightness minimum value in the case of, go bad
Point intensity be MxN data windows in current point with passage brightness minimum value;For current point brightness value in MxN data windows
The situation between brightness maxima and brightness minimum value with current point with passage, removes the brightness value that bad point intensity is current point.
6. the device that dead pixels of image sensor as claimed in claim 1 is surveyed and corrected, it is characterised in that the bad point correction
Unit includes:
One maximum value calculation module, and the gain compensation unit, the bad point judging unit and described goes bad point strength control
Unit connect, for calculate described image sensor data each point MxN data windows in current point with the point of passage
Brightness maxima;
One minimum value computing module, and the gain compensation unit, the bad point judging unit and described goes bad point strength control
Unit connect, for calculate described image sensor data each point MxN data windows in current point with the point of passage
Brightness minimum value;
One mean value computation module, and the gain compensation unit, the bad point judging unit and described removes bad point strength control list
Unit's connection, in the MxN data windows for each point for calculating described image sensor data with current point with the flat of the point of passage
Equal brightness;And
One bad point corrects intensity control module, with the maximum value calculation module, the minimum value computing module, the average meter
Calculate module and it is described go bad point data outputting unit to connect, for according to the brightness maxima, the brightness minimum value, described
Mean flow rate, bad point information and bad point Strength co-mputation is gone to obtain bad point corrected value, and it is described using bad point corrected value correction
Through gain calculating and the vision sensor data of gain compensation.
7. a kind of method that dead pixels of image sensor is surveyed and corrected, it is characterised in that including:
Input picture sensing data;
Data to each passage in the MxN data windows centered on each point of described image sensor data carry out gain
Calculate and gain compensation;
Judge through the vision sensor data of gain calculating and gain compensation each point be in all directions of MxN data windows
It is no to meet bad point condition, and will determine that result as bad point information output;
Calculate described image sensor data each point centered on MxN data windows in current point with the point of passage variance
And average brightness value, the texture strength of current point is calculated using the variance and the mean flow rate, according to the institute being calculated
That states texture strength decision current point goes bad point intensity;
According to the bad point information and it is described go bad point Strength co-mputation to obtain bad point corrected value, and utilize the bad point corrected value school
It is just described through gain calculating and the vision sensor data of gain compensation;And
Vision sensor data after output calibration.
8. the method that dead pixels of image sensor as claimed in claim 7 is surveyed and corrected, it is characterised in that:Described image is sensed
Device data are the data of the various arrangements generation of four filters with 2x2 as cycle, and the form of described image sensor data is
Any one in Bayer format, RGBIR forms, RGBW forms and CyYeMgGr forms.
9. the method that dead pixels of image sensor as claimed in claim 7 is surveyed and corrected, it is characterised in that to described image
The data of each passage in MxN data windows centered on each point of sensing data carry out gain calculating and gain compensation,
Further include:
It is logical as each using the ratio of the luminance mean value of each passage in the MxN data windows and the luminance mean value of first passage
Road needs the gain of compensation, carries out gain calculating;And
Data to each passage are multiplied by the gain that respective channel needs to compensate, and carry out gain compensation.
10. the method that dead pixels of image sensor as claimed in claim 7 is surveyed and corrected, it is characterised in that judge through gain
Calculate with whether each point of the vision sensor data of gain compensation meets bad point in all directions of the MxN data windows
Condition, and will determine that result, as bad point information output, is further included:
Judge through the vision sensor data of gain calculating and gain compensation each point the MxN data windows horizontal direction
On whether meet bad point condition;
Judge through the vision sensor data of gain calculating and gain compensation each point the MxN data windows vertical direction
On whether meet bad point condition;
Judge each point through the vision sensor data of gain calculating and gain compensation in 45 degree of directions of the MxN data windows
On whether meet bad point condition;
Judge each point through the vision sensor data of gain calculating and gain compensation in 135 degree of sides of the MxN data windows
Whether meet bad point condition upwards;And
Judge whether meet in the MxN data simultaneously through each point of the vision sensor data of gain calculating and gain compensation
Four bad point conditions on the horizontal direction of window, vertical direction, 45 degree of directions and 135 degree of directions, will simultaneously meet four bad points
The situation of condition is judged as bad point, and the situation that four bad point conditions are met when will be different is judged as normal point, will be on bad
The judged result of point and normal point is used as bad point information output.
The method that 11. dead pixels of image sensor as claimed in claim 7 are surveyed and corrected, it is characterised in that calculate the figure
As sensing data each point centered on MxN data windows in current point with the point of passage variance and mean flow rate, profit
The texture strength of the current point is calculated with the variance and the mean flow rate, is determined according to the texture strength being calculated
That determines the current point goes bad point intensity, further includes:
Calculate described image sensor data each point centered on MxN data windows in current point with the point of passage side
Difference;
With current point with the average of the point of passage in MxN data windows centered on each point of calculating described image sensor data
Brightness;And
The texture strength of current point is calculated according to the variance and the mean flow rate, and texture is calculated according to formula (a)
Intensity removes bad point intensity W between texture strength lower threshold value and texture strength upper threshold value;
W=(SL-thl)/(thh-thl) (a)
Wherein, SL represents texture strength, and thl represents texture strength lower threshold value, and thh represents texture strength upper threshold value;
When the texture strength is less than texture strength lower threshold value, go bad point intensity in MxN data windows with the same passage of current point
Point mean flow rate;
When the texture strength is more than texture strength upper threshold value, it is more than in MxN data windows and works as the brightness value of current point
Preceding point with the brightness maxima of the point of passage situation, go bad point intensity in MxN data windows with current point with passage brightness
Maximum;For current point brightness value less than in MxN data windows with current point with passage brightness minimum value in the case of, go bad
Point intensity be MxN data windows in current point with passage brightness minimum value;For current point brightness value in MxN data windows
The situation between brightness maxima and brightness minimum value with current point with passage, removes the brightness value that bad point intensity is current point.
The method that 12. dead pixels of image sensor as claimed in claim 7 are surveyed and corrected, it is characterised in that according to described bad
Point information and it is described go bad point Strength co-mputation to obtain bad point corrected value, and correct described through gain meter using the bad point corrected value
The vision sensor data with gain compensation is calculated, is further included:
Calculate described image sensor data each point centered on MxN data windows in current point with the point of passage brightness
Maximum;
Calculate described image sensor data each point centered on MxN data windows in current point with the point of passage brightness
Minimum value;
With current point with the average of the point of passage in MxN data windows centered on each point of calculating described image sensor data
Brightness value;And
According to the brightness maxima, the brightness minimum value, the mean flow rate, bad point information and bad point Strength co-mputation is gone to obtain
It is to bad point corrected value and described through gain calculating and the imageing sensor number of gain compensation using bad point corrected value correction
According to.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510929067.8A CN105430385B (en) | 2015-12-14 | 2015-12-14 | The method and device that a kind of dead pixels of image sensor is surveyed and corrected |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510929067.8A CN105430385B (en) | 2015-12-14 | 2015-12-14 | The method and device that a kind of dead pixels of image sensor is surveyed and corrected |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105430385A CN105430385A (en) | 2016-03-23 |
CN105430385B true CN105430385B (en) | 2017-06-30 |
Family
ID=55508282
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510929067.8A Active CN105430385B (en) | 2015-12-14 | 2015-12-14 | The method and device that a kind of dead pixels of image sensor is surveyed and corrected |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105430385B (en) |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107305695B (en) * | 2016-04-14 | 2021-03-09 | 上海富瀚微电子股份有限公司 | Automatic image dead pixel correction device and method |
CN106210712B (en) * | 2016-08-11 | 2018-07-10 | 上海大学 | A kind of dead pixel points of images detection and processing method |
CN107637068B (en) * | 2016-11-30 | 2019-10-25 | Oppo广东移动通信有限公司 | Dead pixel points of images compensation method, device and electronic equipment |
WO2019196009A1 (en) * | 2018-04-10 | 2019-10-17 | 深圳市大疆创新科技有限公司 | Method for detecting defective pixel of image sensor, and photography apparatus, unmanned aerial vehicle and storage medium |
CN108965749B (en) * | 2018-08-16 | 2021-02-26 | 思特威(上海)电子科技有限公司 | Defect pixel detection and correction device and method based on texture recognition |
CN109348216A (en) * | 2018-11-28 | 2019-02-15 | 凌云光技术集团有限责任公司 | A kind of combination treatment method of bad point detection peace field calibration |
CN109685794B (en) * | 2018-12-25 | 2021-01-29 | 凌云光技术股份有限公司 | Camera self-adaptive step length DPC algorithm and device for mobile phone screen defect detection |
CN110149486B (en) * | 2019-05-17 | 2021-07-30 | 凌云光技术股份有限公司 | Automatic detection method, correction method and system for newly-added abnormal points |
CN110740314B (en) * | 2019-10-30 | 2021-08-10 | 凌云光技术股份有限公司 | Method and system for correcting defective pixel of color line array camera |
CN111083401A (en) * | 2019-12-20 | 2020-04-28 | 成都费恩格尔微电子技术有限公司 | CIS chip dynamic bad point processing method and system |
CN110891172B (en) * | 2019-12-31 | 2021-07-16 | 上海富瀚微电子股份有限公司 | Dead pixel detection and correction device |
CN114095721B (en) * | 2020-08-25 | 2023-04-28 | 京东方科技集团股份有限公司 | Method and device for detecting video dead pixel and computer readable medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6665009B1 (en) * | 1998-05-20 | 2003-12-16 | Omnivision Technologies, Inc. | On-chip dead pixel correction in a CMOS imaging sensor |
CN101715050A (en) * | 2009-11-20 | 2010-05-26 | 北京中星微电子有限公司 | Method and system for detecting dead pixels of image sensor |
CN102640489A (en) * | 2009-10-20 | 2012-08-15 | 苹果公司 | System and method for detecting and correcting defective pixels in an image sensor |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20150014556A (en) * | 2013-07-29 | 2015-02-09 | 에스케이하이닉스 주식회사 | Apparatus for correcting dead pixel and method thereof, and cmos image sensor using that |
-
2015
- 2015-12-14 CN CN201510929067.8A patent/CN105430385B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6665009B1 (en) * | 1998-05-20 | 2003-12-16 | Omnivision Technologies, Inc. | On-chip dead pixel correction in a CMOS imaging sensor |
CN102640489A (en) * | 2009-10-20 | 2012-08-15 | 苹果公司 | System and method for detecting and correcting defective pixels in an image sensor |
CN101715050A (en) * | 2009-11-20 | 2010-05-26 | 北京中星微电子有限公司 | Method and system for detecting dead pixels of image sensor |
Also Published As
Publication number | Publication date |
---|---|
CN105430385A (en) | 2016-03-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105430385B (en) | The method and device that a kind of dead pixels of image sensor is surveyed and corrected | |
CN106210712B (en) | A kind of dead pixel points of images detection and processing method | |
CN102202163B (en) | Adaptive enhancement method and device for monitored video | |
CN110248112A (en) | A kind of exposal control method of imaging sensor | |
DE112012004641T5 (en) | Vehicle vision system with color correction | |
US20040041928A1 (en) | Image processing apparatus, camera apparatus, and automatic exposure control method | |
CN102724405A (en) | Method and device for automatic exposure compensation of backlit scenes in video imaging system | |
WO2006129492A1 (en) | Vehicle and road sign recognition device | |
CN105578021A (en) | Imaging method of binocular camera and apparatus thereof | |
US10326969B2 (en) | Vehicle vision system with reduction of temporal noise in images | |
CN106531049A (en) | Brightness adjusting method and system of display panel | |
CN102447815B (en) | Video image processing method and device | |
CN103714520A (en) | Digital video image enhancement achieving system and method based on FPGA | |
CN103460682A (en) | Image processing device and method | |
CN101068312A (en) | Exposure compensation method for digital image | |
CN106713790A (en) | Control method, control device and electronic device | |
CN103327219A (en) | Video image contrast adjustment method and device | |
CN104112118B (en) | Method for detecting lane lines for Lane Departure Warning System | |
CN112861645A (en) | Infrared camera dim light environment compensation method and device and electronic equipment | |
CN104079909A (en) | Automatic white balance method and device for image | |
CN106500855A (en) | A kind of Infrared Detectorss blind pixel detection method | |
CN106134180A (en) | The record medium of image processing program that image processing apparatus and method, storage can be read by computer temporarily, camera head | |
CN107292387A (en) | A kind of method that honey quality is recognized based on BP | |
CN105530406B (en) | A kind of intelligent Enhancement Method based on historical frames image statistics value | |
CN100481903C (en) | Adaptive LCD TV display quality adjusting apparatus and method |
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 |