CN110264415A - It is a kind of to eliminate the fuzzy image processing method of shake - Google Patents
It is a kind of to eliminate the fuzzy image processing method of shake Download PDFInfo
- Publication number
- CN110264415A CN110264415A CN201910442125.2A CN201910442125A CN110264415A CN 110264415 A CN110264415 A CN 110264415A CN 201910442125 A CN201910442125 A CN 201910442125A CN 110264415 A CN110264415 A CN 110264415A
- Authority
- CN
- China
- Prior art keywords
- image
- fuzzy
- pixel
- shake
- described image
- 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.)
- Granted
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 30
- 238000006243 chemical reaction Methods 0.000 claims abstract description 50
- 238000013135 deep learning Methods 0.000 claims abstract description 37
- 238000011946 reduction process Methods 0.000 claims abstract description 19
- 230000008030 elimination Effects 0.000 claims abstract description 10
- 238000003379 elimination reaction Methods 0.000 claims abstract description 10
- 238000000034 method Methods 0.000 claims description 18
- 238000001914 filtration Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 10
- 238000001228 spectrum Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 8
- 239000003086 colorant Substances 0.000 claims description 3
- 239000012634 fragment Substances 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 239000004744 fabric Substances 0.000 claims description 2
- 238000004458 analytical method Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000006641 stabilisation Effects 0.000 description 4
- 238000011105 stabilization Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000003707 image sharpening Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
Classifications
-
- G06T5/70—
-
- G06T5/73—
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
- H04N23/80—Camera processing pipelines; Components thereof
Abstract
The fuzzy image processing method of shake is eliminated the present invention provides a kind of, the fuzzy image processing method of elimination shake is by successively carrying out noise reduction process and Edge contrast to image, and the result based on the Edge contrast, determine the corresponding fuzzy message of the image, it is based on the fuzzy message again, pass through the mode of deep learning, the first image operation composed about image processing is carried out to the image, intermediate conversion image is obtained with this, finally by the mode of deep learning, the intermediate conversion image handle about the second image operation of image resolution ratio, the shake eliminated in the image is fuzzy, the sharpening processing to the image is realized with this.
Description
Technical field
The present invention relates to the technical fields of image procossing, in particular to a kind of to eliminate the fuzzy image processing method of shake.
Background technique
With popularizing for smart phone, smart phone has been dissolved into people's lives and work, and people pass through intelligence
Mobile phone can be realized various forms of operations such as instant messaging, web page browsing, amusement and light office.Smart phone why can
Huge development is obtained, exactly smart phone combines traditional mobile phone with the different function module such as camera, thus
So that smart phone is no longer limited to traditional call function, and other intelligent operations being differently formed can also be carried out.Intelligence
The camera function of camera can constantly it enhance in mobile phone, image taking and in time can be carried out whenever and wherever possible by being also convenient for people
Ground is shared by the image taken by cloud.
It is by people always in image taking although the camera of smart phone has powerful image camera function
Hand operation is completed, this is resulted in shooting process, since the influence of external environment or the oneself factor of photographer cause
The hand of shooting is shaken, so that camera, there is a situation where defocus, the image correspondingly taken also necessarily occurs
Smudgy situation.Since the case where hand shake is widely present and unpredictable and artificially avoid, this claps image
This uncertain factor is obscured in the presence of shake always during taking the photograph.In order to avoid hand shake causes the image of shooting to generate mould
Paste, the prior art in the imaging lens system of camera by being arranged stabilization lens unit, the stabilization lens unit energy
Enough to move under the driving of control signal, the movement of the stabilization lens unit can effectively cancel out hand shake and be led
The defocus situation of cause, although this stabilization dynamic component by the way that example, in hardware is arranged in camera can be effectively reduced hand
Image caused by shaking is fuzzy, but this mode increases the volume and hardware cost of camera, while it is to control signal
Adjustment also proposed higher requirement, this for pursue miniaturization smart phone for there will naturally be conflict place, and
It also proposed harsher requirement for the control performance of smart phone.As it can be seen that the prior art is schemed caused by shaking for hand
As fuzzy processing mode is realized by way of hardware, there is no overcome in such a way that software is handled at present
Blurred image technology caused by hand is shaken.
Summary of the invention
In view of the defects existing in the prior art, the present invention provides a kind of image processing method eliminating shake and obscuring, this disappears
Except the fuzzy image processing method of shake is by successively carrying out noise reduction process and Edge contrast to image, and based at the sharpening
Reason as a result, determine the corresponding fuzzy message of the image, then the fuzzy message is based on, by the mode of deep learning, to the figure
As the first image operation processing compose about image, intermediate conversion image is obtained with this, finally by the mould of deep learning
Formula handle about the second image operation of image resolution ratio to the intermediate conversion image, eliminates the shake mould in the image
Paste realizes the sharpening processing to the image with this.The fuzzy image processing method of elimination shake is different from the prior art
The mode of dependence hardware device obscures to reduce image caused by hand is shaken, the image which obtains shooting
Using the form of subsequent calculation processing, the analysis in algorithm level is carried out to image and is handled, analysis processing does not need any
The support of additional hardware equipment only needs after the completion of image taking, carries out at corresponding noise reduction process and sharpening to image
Reason is obtained corresponding image fuzzy message with this, then is based on the image fuzzy message, and it is deep to carry out corresponding neural network to the image
Study processing is spent, to eliminate the fuzzy pixel in the image, and finally realizes the sharpening of image.
The present invention provides a kind of image processing method eliminating shake and obscuring, which is characterized in that the elimination shake is fuzzy
Image processing method include the following steps:
Step (1) successively carries out noise reduction process and Edge contrast to image；
Step (2), based on the Edge contrast as a result, determining the corresponding fuzzy message of described image；
Step (3) is based on the fuzzy message, by the mode of deep learning, compose about image to described image
The first image operation processing, intermediate conversion image is obtained with this；
Step (4) carries out second about image resolution ratio to the intermediate conversion image by the mode of deep learning
Image operation processing, the shake eliminated in described image is fuzzy, realizes the sharpening processing to described image with this；
Further, in the step (1), noise reduction process is carried out to described image and is specifically included,
Step (A101) obtains the Luminance Distribution information of described image, the Luminance Distribution information is based on, by described image
It is divided into several different texture regions with different images grain distribution state；
Step (A102) calculates and obtains the corresponding brightness point of each of several different texture regions texture region
Amount and/or texture component, and according to the luminance component and/or the texture component by all pixels point in described image
It is divided into first kind pixel and Second Type pixel；
Step (A103) executes the filter of adaptability to the first kind pixel and the Second Type pixel respectively
Wave noise reduction process；
Further, in the step (A101), the Luminance Distribution information of described image is obtained, is based on the Luminance Distribution
Described image is divided into several different texture regions with different images grain distribution state and specifically included by information,
Step (A1011) obtains the Luminance Distribution information and tone distributed intelligence of described image, according to the Luminance Distribution
Information and the tone distributed intelligence are calculated about described image in the corresponding brightness in whole image region-tone related information；
Step (A1012) determines the figure according in the corresponding brightness-tone related information in whole image region
As in the corresponding image texture distribution in whole image region, wherein described image grain distribution state include at least about
The light and shade fringe distribution information of image and/or three primary colours distributed intelligence about image；
Described image is divided into several different texture areas according to described image grain distribution state by step (A1013)
Domain；
Further, in the step (A102), according to the luminance component and/or the texture component by described image
In all pixels point be divided into first kind pixel and Second Type pixel specifically includes,
Step (A1021) calculates each pixel in described image according to the luminance component and the texture component
The brightness-interference of texture coefficient and default interference coefficient range are compared place by corresponding brightness-interference of texture coefficient
Reason；
Step (A1022), if the brightness-interference of texture coefficient is within the scope of the default interference coefficient, it is determined that
Corresponding pixel is first kind pixel, if the brightness-interference of texture coefficient is not at the default interference coefficient model
In enclosing, it is determined that corresponding pixel fragment is Second Type pixel, wherein the first kind pixel is noise type pixel
Point, the Second Type pixel are non-noise classes of pixels point；
Alternatively,
In the step (A103), the first kind pixel and the Second Type pixel are executed respectively suitable
The filtering noise reduction process of answering property specifically includes,
Step (A1031) executes gaussian filtering process or karr to full pixel area in the first kind pixel
Graceful filtering processing, so that the noise coefficient of the first kind pixel meets preset condition；
Step (A1032) executes smoothing filtering processing to the high-frequency pixels region in the Second Type pixel, with
The noise coefficient of the Second Type pixel is set to meet preset condition；
Further, in the step (1), processing is sharpened to described image and is specifically included,
Step (B101) carries out image mould to the noise-reduced image obtained after the noise reduction process by Sobel operator
Gelatinization conversion；
Step (B102) determines the fringe region of the blurred image after described image blurring conversion, wherein described
Fringe region includes at least the corresponding region of pre-determined distance range inside and outside the boundary and the boundary of the blurred image；
Step (B103) obtains about the corresponding fuzzy characteristics of the fringe region, the marginal zone is calculated with this
The corresponding fuzzy core in domain, wherein the fuzzy characteristics includes at least one of blur direction, blurring trajectorie and fuzzy angle；
Further, in the step (2), based on the Edge contrast as a result, determining that described image is corresponding fuzzy
Information specifically includes,
Step (201) determines described image based on the fuzzy nuclear information about described image that the Edge contrast obtains
Region-wide Fuzzy Correlation coefficient；
Step (202), based on the distribution of the region-wide Fuzzy Correlation coefficient of described image, to the whole district of described image
Domain carries out the division in several different fog-levels regions；
Step (203) calculates each fog-level area based on the division result in several different fog-levels regions
The corresponding pixel Fuzzy Distribution tendency information in domain, in this, as the fuzzy message；
Further, in the step (3), it is based on the fuzzy message, by the mode of deep learning, to described image
The first image operation composed about image processing is carried out, intermediate conversion image is obtained with this and is specifically included,
Step (301), based on the fuzzy message about described image the corresponding fuzzy vector of different zones pixel,
At least one of blur direction, blurring trajectorie and fuzzy angle are carried out Pixel fit conversion process to described image, are obtained with this
Obtain corresponding Pixel fit image；
Step (302) carries out the Pixel fit image relevant to image spectrum by the mode of deep learning
The processing of first image operation, and the intermediate conversion image is calculated；
Further, in the step (302), by the mode of deep learning, the Pixel fit image is carried out
The first image operation processing relevant to image spectrum, and the intermediate conversion image is calculated and specifically includes,
By the mode of deep learning, least square method relevant to image spectrum is carried out to the Pixel fit image
Image deconvolution operation corresponding with Landweber iterative method, so that the intermediate conversion image be calculated；
Further, in the step (4), by the mode of deep learning, to the intermediate conversion image carry out about
Second image operation of image resolution ratio is handled, and the shake in elimination described image is fuzzy to be specifically included,
Step (401) obtains the intermediate conversion image in region-wide resolution distribution state, and judges the resolution
Whether rate distribution meets default resolution distribution condition；
Step (402), if the resolution distribution state meets the default resolution distribution condition, to the centre
Changing image carries out the second image operation processing, if the resolution distribution state does not meet the default resolution distribution
Condition then carries out fuzzy pixel rejecting processing to the intermediate conversion image, until the resolution distribution state meets institute
Until stating default resolution distribution condition；
Step (403), then by the mode of deep learning, to the intermediate conversion for meeting the default rate distribution occasion respectively
Image executes the second image operation processing, fuzzy to eliminate the shake in described image；
Further, in the step (403), the intermediate conversion image for meeting the default rate distribution occasion respectively is held
Row the second image operation processing specifically includes,
By the mode of deep learning, all pixels of the intermediate conversion image are performed both by with the meter of Super-Resolution
Calculation processing realizes that all pixels are restored in the promotion of at least one of brightness value, chromatic value and tone value with this, realizes institute with this
State that sharpening is handled and to eliminate the shake in described image fuzzy.
Compared with the prior art, the fuzzy image processing method of elimination shake is by successively carrying out noise reduction process to image
And Edge contrast, and based on the Edge contrast as a result, determine the corresponding fuzzy message of the image, then based on the fuzzy letter
Breath carries out the first image operation composed about image to the image and handles, obtain intermediate change with this by the mode of deep learning
Image is changed, finally by the mode of deep learning, which transport about the second image of image resolution ratio
Calculation processing, the shake eliminated in the image is fuzzy, realizes the sharpening processing to the image with this.The fuzzy figure of elimination shake
As processing method be different from the mode of the dependence hardware device of the prior art reduce hand shake caused by image obscure, the figure
As processing method uses the form of subsequent calculation processing to the obtained image of shooting, image is carried out at the analysis in algorithm level
Reason, which, which is handled, does not need the support of any additional hardware equipment, only needs after the completion of image taking, to image into
The corresponding noise reduction process of row and Edge contrast are obtained corresponding image fuzzy message with this, then are based on the image fuzzy message, right
The image carries out corresponding neural network deep learning processing, to eliminate the fuzzy pixel in the image, and finally realizes image
Sharpening.
Other features and advantages of the present invention will be illustrated in the following description, also, partly becomes from specification
It obtains it is clear that understand through the implementation of the invention.The objectives and other advantages of the invention can be by written explanation
Specifically noted structure is achieved and obtained in book, claims and attached drawing.
Below by drawings and examples, technical scheme of the present invention will be described in further detail.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of flow diagram for eliminating the fuzzy image processing method of shake provided by the invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
It refering to fig. 1, is a kind of process signal for eliminating the fuzzy image processing method of shake provided in an embodiment of the present invention
Figure.The fuzzy image processing method of elimination shake includes the following steps:
Step (1) successively carries out noise reduction process and Edge contrast to image.
Preferably, in the step (1), noise reduction process is carried out to the image and is specifically included,
Step (A101) obtains the Luminance Distribution information of the image, is based on the Luminance Distribution information, divides the image into
Several different texture regions with different images grain distribution state；
Step (A102) calculates and obtains each of several different texture regions corresponding luminance component of texture region
And/or texture component, and all pixels point in the image is divided into according to the luminance component and/or the texture component
One type pixel and Second Type pixel；
Step (A103), the filtering for executing adaptability respectively to the first kind pixel and the Second Type pixel are dropped
It makes an uproar processing.
Preferably, in the step (A101), the Luminance Distribution information of the image is obtained, is based on the Luminance Distribution information,
Several different texture regions with different images grain distribution state are divided the image into specifically include,
Step (A1011) obtains the Luminance Distribution information and tone distributed intelligence of the image, according to the Luminance Distribution information
With the tone distributed intelligence, calculate about the image in the corresponding brightness in whole image region-tone related information；
Step (A1012) determines that the image exists according in the corresponding brightness-tone related information in whole image region
The corresponding image texture distribution in whole image region, wherein the image texture distribution is included at least about image
Light and shade fringe distribution information and/or three primary colours distributed intelligence about image；
Step (A1013) divides the image into several different texture regions according to the image texture distribution.
It preferably, will be all in the image according to the luminance component and/or the texture component in the step (A102)
Pixel is divided into first kind pixel and Second Type pixel specifically includes,
Step (A1021), according to the luminance component and the texture component, each pixel calculated in the image is corresponding
The brightness-interference of texture coefficient and default interference coefficient range are compared processing by brightness-interference of texture coefficient；
Step (A1022), if the brightness-interference of texture coefficient is within the scope of the default interference coefficient, it is determined that corresponding
Pixel be first kind pixel, if the brightness-interference of texture coefficient is not within the scope of the default interference coefficient, really
Fixed corresponding pixel fragment is Second Type pixel, wherein the first kind pixel is noise type pixel, second class
Type pixel is non-noise classes of pixels point.
Preferably, in the step (A103), the first kind pixel and the Second Type pixel are executed respectively
The filtering noise reduction process of adaptability specifically includes,
Step (A1031) executes gaussian filtering process or Kalman to full pixel area in the first kind pixel
Filtering processing, so that the noise coefficient of the first kind pixel meets preset condition；
Step (A1032) executes smoothing filtering processing to the high-frequency pixels region in the Second Type pixel, so that
The noise coefficient of the Second Type pixel meets preset condition.
Preferably, in the step (1), processing is sharpened to the image and is specifically included,
It is fuzzy to carry out image to the noise-reduced image obtained after the noise reduction process by Sobel operator for step (B101)
Change conversion；
Step (B102) determines the fringe region of the blurred image after image blurring conversion, wherein the edge
Region includes at least the corresponding region of pre-determined distance range inside and outside the boundary and the boundary of the blurred image；
Step (B103) obtains about the corresponding fuzzy characteristics of the fringe region, the fringe region pair is calculated with this
The fuzzy core answered, wherein the fuzzy characteristics includes at least one of blur direction, blurring trajectorie and fuzzy angle.
Step (2), based on the Edge contrast as a result, determining the corresponding fuzzy message of the image.
Preferably, in the step (2), based on the Edge contrast as a result, determining that the corresponding fuzzy message of the image has
Body includes,
Step (201) determines image entire area based on the fuzzy nuclear information about the image that the Edge contrast obtains
Fuzzy Correlation coefficient；
Step (202), based on the distribution of the region-wide Fuzzy Correlation coefficient of the image, to the image it is region-wide into
The division in several different fog-levels regions of row；
Step (203) calculates each fog-level region based on the division result in several different fog-levels regions
Corresponding pixel Fuzzy Distribution tendency information, in this, as the fuzzy message.
Step (3) is based on the fuzzy message, by the mode of deep learning, which compose about image the
The processing of one image operation, obtains intermediate conversion image with this.
Preferably, in the step (3), the image is closed by the mode of deep learning based on the fuzzy message
In the first image operation processing of image spectrum, intermediate conversion image is obtained with this and is specifically included,
Step (301), the corresponding fuzzy vector of different zones pixel, fuzzy about the image based on the fuzzy message
At least one of direction, blurring trajectorie and fuzzy angle carry out Pixel fit conversion process to the image, obtain correspondence with this
Pixel fit image；
Step (302) carries out relevant to image spectrum the to the Pixel fit image by the mode of deep learning
The processing of one image operation, and the intermediate conversion image is calculated.
Preferably, in the step (302), by the mode of deep learning, which is carried out and is schemed
Picture frequency composes relevant first image operation processing, and the intermediate conversion image is calculated and specifically includes,
By the mode of deep learning, to the Pixel fit image carry out least square method relevant to image spectrum and
The corresponding image deconvolution operation of Landweber iterative method, so that the intermediate conversion image be calculated.
Step (4) carries out the second figure about image resolution ratio to the intermediate conversion image by the mode of deep learning
As calculation process, the shake eliminated in the image is fuzzy, realizes the sharpening processing to the image with this.
Preferably, in the step (4), by the mode of deep learning, which is carried out about image
Second image operation of resolution ratio is handled, and eliminates that the shake in the image is fuzzy to be specifically included,
Step (401) obtains the intermediate conversion image in region-wide resolution distribution state, and judges the resolution ratio point
Whether cloth state meets default resolution distribution condition；
Step (402), if the resolution distribution state meets the default resolution distribution condition, to the intermediate conversion figure
As carrying out second image operation processing, if the resolution distribution state does not meet the default resolution distribution condition, to this
Intermediate conversion image carries out fuzzy pixel rejecting processing, until the resolution distribution state meets the default resolution distribution item
Until part；
Step (403), then by the mode of deep learning, to the intermediate conversion figure for meeting the default rate distribution occasion respectively
It is fuzzy to eliminate the shake in the image as executing second image operation processing.
Preferably, in the step (403), executing to the intermediate conversion image for meeting the default rate distribution occasion respectively should
The processing of second image operation specifically includes,
By the mode of deep learning, all pixels of the intermediate conversion image are performed both by with the calculating of Super-Resolution
Processing realizes that all pixels are restored in the promotion of at least one of brightness value, chromatic value and tone value with this, realizes that this is clear with this
Clearization handles and eliminates the shake in the image and obscure.
From above-described embodiment as it can be seen that the elimination shakes fuzzy image processing method by successively carrying out from noise reduction to image
Reason and Edge contrast, and based on the Edge contrast as a result, determine the corresponding fuzzy message of the image, then based on the fuzzy letter
Breath carries out the first image operation composed about image to the image and handles, obtain intermediate change with this by the mode of deep learning
Image is changed, finally by the mode of deep learning, which transport about the second image of image resolution ratio
Calculation processing, the shake eliminated in the image is fuzzy, realizes the sharpening processing to the image with this.The fuzzy figure of elimination shake
As processing method be different from the mode of the dependence hardware device of the prior art reduce hand shake caused by image obscure, the figure
As processing method uses the form of subsequent calculation processing to the obtained image of shooting, image is carried out at the analysis in algorithm level
Reason, which, which is handled, does not need the support of any additional hardware equipment, only needs after the completion of image taking, to image into
The corresponding noise reduction process of row and Edge contrast are obtained corresponding image fuzzy message with this, then are based on the image fuzzy message, right
The image carries out corresponding neural network deep learning processing, to eliminate the fuzzy pixel in the image, and finally realizes image
Sharpening.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (10)
1. a kind of eliminate the fuzzy image processing method of shake, which is characterized in that described to eliminate the fuzzy image processing method of shake
Method includes the following steps:
Step (1) successively carries out noise reduction process and Edge contrast to image；
Step (2), based on the Edge contrast as a result, determining the corresponding fuzzy message of described image；
Step (3) is based on the fuzzy message, by the mode of deep learning, described image compose about image the
The processing of one image operation, obtains intermediate conversion image with this；
Step (4) carries out the second image about image resolution ratio to the intermediate conversion image by the mode of deep learning
Calculation process, the shake eliminated in described image is fuzzy, realizes the sharpening processing to described image with this.
2. as described in claim 1 eliminate the fuzzy image processing method of shake, it is characterised in that:
In the step (1), noise reduction process is carried out to described image and is specifically included,
Step (A101) obtains the Luminance Distribution information of described image, is based on the Luminance Distribution information, described image is divided
For several different texture regions with different images grain distribution state；
Step (A102) calculates and obtains each of several different texture regions corresponding luminance component of texture region
And/or texture component, and according to the luminance component and/or the texture component by all pixels click and sweep in described image
It is divided into first kind pixel and Second Type pixel；
Step (A103), the filtering for executing adaptability respectively to the first kind pixel and the Second Type pixel are dropped
It makes an uproar processing.
3. as claimed in claim 2 eliminate the fuzzy image processing method of shake, it is characterised in that:
In the step (A101), the Luminance Distribution information of described image is obtained, is based on the Luminance Distribution information, it will be described
Image is divided into several different texture regions with different images grain distribution state and specifically includes,
Step (A1011) obtains the Luminance Distribution information and tone distributed intelligence of described image, according to the Luminance Distribution information
With the tone distributed intelligence, calculate about described image in the corresponding brightness in whole image region-tone related information；
Step (A1012) determines that described image exists according in the corresponding brightness-tone related information in whole image region
The corresponding image texture distribution in whole image region, wherein described image grain distribution state is included at least about image
Light and shade fringe distribution information and/or three primary colours distributed intelligence about image；
Described image is divided into several different texture regions according to described image grain distribution state by step (A1013).
4. as claimed in claim 2 eliminate the fuzzy image processing method of shake, it is characterised in that:
In the step (A102), according to the luminance component and/or the texture component by all pictures in described image
Vegetarian refreshments is divided into first kind pixel and Second Type pixel specifically includes,
Step (A1021), according to the luminance component and the texture component, each pixel calculated in described image is corresponding
Brightness-interference of texture coefficient, the brightness-interference of texture coefficient and default interference coefficient range are compared into processing；
Step (A1022), if the brightness-interference of texture coefficient is within the scope of the default interference coefficient, it is determined that corresponding
Pixel be first kind pixel, if the brightness-interference of texture coefficient is not within the scope of the default interference coefficient,
Then determine that corresponding pixel fragment is Second Type pixel, wherein the first kind pixel is noise type pixel, institute
Stating Second Type pixel is non-noise classes of pixels point；
Alternatively,
In the step (A103), adaptability is executed respectively to the first kind pixel and the Second Type pixel
Filtering noise reduction process specifically include,
Step (A1031) executes gaussian filtering process to full pixel area in the first kind pixel or Kalman filters
Wave processing, so that the noise coefficient of the first kind pixel meets preset condition；
Step (A1032) executes smoothing filtering processing to the high-frequency pixels region in the Second Type pixel, so that institute
The noise coefficient for stating Second Type pixel meets preset condition.
5. as described in claim 1 eliminate the fuzzy image processing method of shake, it is characterised in that:
In the step (1), processing is sharpened to described image and is specifically included,
Step (B101) carries out image blurring to the noise-reduced image obtained after the noise reduction process by Sobel operator
Conversion；
Step (B102) determines the fringe region of the blurred image after described image blurring conversion, wherein the edge
Region includes at least the corresponding region of pre-determined distance range inside and outside the boundary and the boundary of the blurred image；
Step (B103) obtains about the corresponding fuzzy characteristics of the fringe region, the fringe region pair is calculated with this
The fuzzy core answered, wherein the fuzzy characteristics includes at least one of blur direction, blurring trajectorie and fuzzy angle.
6. as described in claim 1 eliminate the fuzzy image processing method of shake, it is characterised in that:
In the step (2), based on the Edge contrast as a result, determining that the corresponding fuzzy message of described image is specifically wrapped
It includes,
Step (201) determines the described image whole district based on the fuzzy nuclear information about described image that the Edge contrast obtains
The Fuzzy Correlation coefficient in domain；
Step (202), based on the distribution of the region-wide Fuzzy Correlation coefficient of described image, to described image it is region-wide into
The division in several different fog-levels regions of row；
It is each to calculate each fog-level region based on the division result in several different fog-levels regions for step (203)
Self-corresponding pixel Fuzzy Distribution tendency information, in this, as the fuzzy message.
7. as described in claim 1 eliminate the fuzzy image processing method of shake, it is characterised in that:
In the step (3), it is based on the fuzzy message, by the mode of deep learning, described image is carried out about figure
As the first image operation processing of spectrum, intermediate conversion image is obtained with this and is specifically included,
Step (301), the corresponding fuzzy vector of different zones pixel, fuzzy about described image based on the fuzzy message
At least one of direction, blurring trajectorie and fuzzy angle are carried out Pixel fit conversion process to described image, are obtained pair with this
The Pixel fit image answered；
Step (302) carries out relevant to image spectrum first to the Pixel fit image by the mode of deep learning
Image operation processing, and the intermediate conversion image is calculated.
8. as claimed in claim 7 eliminate the fuzzy image processing method of shake, it is characterised in that:
In the step (302), by the mode of deep learning, the Pixel fit image is carried out and image spectrum phase
The the first image operation processing closed, and the intermediate conversion image is calculated and specifically includes,
By the mode of deep learning, to the Pixel fit image carry out least square method relevant to image spectrum and
The corresponding image deconvolution operation of Landweber iterative method, so that the intermediate conversion image be calculated.
9. as described in claim 1 eliminate the fuzzy image processing method of shake, it is characterised in that:
In the step (4), by the mode of deep learning, the intermediate conversion image is carried out about image resolution ratio
Second image operation is handled, and the shake in elimination described image is fuzzy to be specifically included,
Step (401) obtains the intermediate conversion image in region-wide resolution distribution state, and judges the resolution ratio point
Whether cloth state meets default resolution distribution condition；
Step (402), if the resolution distribution state meets the default resolution distribution condition, to the intermediate conversion
Image carries out the second image operation processing, if the resolution distribution state does not meet the default resolution distribution item
Part then carries out fuzzy pixel rejecting processing to the intermediate conversion image, until the resolution distribution state meet it is described
Until default resolution distribution condition；
Step (403), then by the mode of deep learning, to the intermediate conversion image for meeting the default rate distribution occasion respectively
The second image operation processing is executed, it is fuzzy to eliminate the shake in described image.
10. as claimed in claim 9 eliminate the fuzzy image processing method of shake, it is characterised in that:
In the step (403), described second is executed to the intermediate conversion image for meeting the default rate distribution occasion respectively
Image operation processing specifically includes,
By the mode of deep learning, all pixels of the intermediate conversion image are performed both by the calculating of Super-Resolution
Reason is realized that all pixels are restored in the promotion of at least one of brightness value, chromatic value and tone value with this, is realized with this described clear
Clearization handles and eliminates the shake in described image and obscure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910442125.2A CN110264415B (en) | 2019-05-24 | 2019-05-24 | Image processing method for eliminating jitter blur |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910442125.2A CN110264415B (en) | 2019-05-24 | 2019-05-24 | Image processing method for eliminating jitter blur |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110264415A true CN110264415A (en) | 2019-09-20 |
| CN110264415B CN110264415B (en) | 2020-06-12 |
Family
ID=67915390
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910442125.2A Active CN110264415B (en) | 2019-05-24 | 2019-05-24 | Image processing method for eliminating jitter blur |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110264415B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111340735A (en) * | 2020-03-04 | 2020-06-26 | 卡莱特（深圳）云科技有限公司 | LED screen body correction method, device and terminal |
| CN112019749A (en) * | 2020-09-02 | 2020-12-01 | 中国银行股份有限公司 | Camera adjusting method and device |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101213574A (en) * | 2005-06-07 | 2008-07-02 | 汤姆逊许可公司 | Content-based Gaussian noise reduction for still image, video and film |
| CN102184533A (en) * | 2011-06-10 | 2011-09-14 | 西安电子科技大学 | Non-local-restriction-based total variation image deblurring method |
| CN104616257A (en) * | 2015-01-26 | 2015-05-13 | 山东省计算中心（国家超级计算济南中心） | Recovery evidence obtaining method for blurred degraded digital images in administration of justice |
| CN106447626A (en) * | 2016-09-07 | 2017-02-22 | 华中科技大学 | Blurred kernel dimension estimation method and system based on deep learning |
| CN106530256A (en) * | 2016-11-18 | 2017-03-22 | 四川长虹电器股份有限公司 | Improved-deep-learning-based intelligent camera image blind super-resolution system |
| CN107730453A (en) * | 2017-11-13 | 2018-02-23 | 携程计算机技术（上海）有限公司 | Picture quality method for improving |
| CN107945125A (en) * | 2017-11-17 | 2018-04-20 | 福州大学 | It is a kind of to merge spectrum estimation method and the fuzzy image processing method of convolutional neural networks |
| CN108389163A (en) * | 2018-01-22 | 2018-08-10 | 西安工程大学 | A kind of self-adapting enhancement method based on half-light coloured image |
| CN108765511A (en) * | 2018-05-30 | 2018-11-06 | 重庆大学 | Ultrasonoscopy super resolution ratio reconstruction method based on deep learning |
| CN109145796A (en) * | 2018-08-13 | 2019-01-04 | 福建和盛高科技产业有限公司 | A kind of identification of electric power piping lane fire source and fire point distance measuring method based on video image convergence analysis algorithm |
| CN109377459A (en) * | 2018-09-30 | 2019-02-22 | 国网山东省电力公司电力科学研究院 | A kind of super-resolution deblurring method of production confrontation network |
-
2019
- 2019-05-24 CN CN201910442125.2A patent/CN110264415B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101213574A (en) * | 2005-06-07 | 2008-07-02 | 汤姆逊许可公司 | Content-based Gaussian noise reduction for still image, video and film |
| CN102184533A (en) * | 2011-06-10 | 2011-09-14 | 西安电子科技大学 | Non-local-restriction-based total variation image deblurring method |
| CN104616257A (en) * | 2015-01-26 | 2015-05-13 | 山东省计算中心（国家超级计算济南中心） | Recovery evidence obtaining method for blurred degraded digital images in administration of justice |
| CN106447626A (en) * | 2016-09-07 | 2017-02-22 | 华中科技大学 | Blurred kernel dimension estimation method and system based on deep learning |
| CN106530256A (en) * | 2016-11-18 | 2017-03-22 | 四川长虹电器股份有限公司 | Improved-deep-learning-based intelligent camera image blind super-resolution system |
| CN107730453A (en) * | 2017-11-13 | 2018-02-23 | 携程计算机技术（上海）有限公司 | Picture quality method for improving |
| CN107945125A (en) * | 2017-11-17 | 2018-04-20 | 福州大学 | It is a kind of to merge spectrum estimation method and the fuzzy image processing method of convolutional neural networks |
| CN108389163A (en) * | 2018-01-22 | 2018-08-10 | 西安工程大学 | A kind of self-adapting enhancement method based on half-light coloured image |
| CN108765511A (en) * | 2018-05-30 | 2018-11-06 | 重庆大学 | Ultrasonoscopy super resolution ratio reconstruction method based on deep learning |
| CN109145796A (en) * | 2018-08-13 | 2019-01-04 | 福建和盛高科技产业有限公司 | A kind of identification of electric power piping lane fire source and fire point distance measuring method based on video image convergence analysis algorithm |
| CN109377459A (en) * | 2018-09-30 | 2019-02-22 | 国网山东省电力公司电力科学研究院 | A kind of super-resolution deblurring method of production confrontation network |
Non-Patent Citations (2)
| Title |
|---|
| XU L, REN J S J, LIU C, ET AL.: ""Deep convolutional neural network for image deconvolution"", 《PROCEEDINGS OF ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS》 * |
| 彭青建: ""运动模糊图像复原算法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111340735A (en) * | 2020-03-04 | 2020-06-26 | 卡莱特（深圳）云科技有限公司 | LED screen body correction method, device and terminal |
| CN111340735B (en) * | 2020-03-04 | 2021-04-23 | 卡莱特云科技股份有限公司 | LED screen body correction method, device and terminal |
| CN112019749A (en) * | 2020-09-02 | 2020-12-01 | 中国银行股份有限公司 | Camera adjusting method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110264415B (en) | 2020-06-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10410327B2 (en) | Shallow depth of field rendering | |
| US9998666B2 (en) | Systems and methods for burst image deblurring | |
| US8509559B2 (en) | Hierarchical motion deblurring method for single image | |
| CN111353948B (en) | Image noise reduction method, device and equipment | |
| US7983511B1 (en) | Methods and apparatus for noise reduction in digital images | |
| US20160301868A1 (en) | Automated generation of panning shots | |
| Delbracio et al. | Removing camera shake via weighted fourier burst accumulation | |
| CN107948500A (en) | Image processing method and device | |
| JP2013513318A (en) | Digital image composition to generate optical effects | |
| CN107430762B (en) | Digital zooming method and system | |
| US10992845B1 (en) | Highlight recovery techniques for shallow depth of field rendering | |
| KR20120027712A (en) | Method and apparatus for processing image | |
| CN105453540B (en) | Image processing apparatus, photographic device, image processing method | |
| US8488899B2 (en) | Image processing apparatus, method and recording medium | |
| US10083500B2 (en) | Restoration filter generation device and method, image processing device and method, imaging device, and non-transitory computer-readable medium | |
| WO2023273868A1 (en) | Image denoising method and apparatus, terminal, and storage medium | |
| CN110022430A (en) | Image weakening method, device, mobile terminal and computer readable storage medium | |
| JP2023538020A (en) | Image processing based on object category classification | |
| CN110264415A (en) | It is a kind of to eliminate the fuzzy image processing method of shake | |
| US9466007B2 (en) | Method and device for image processing | |
| CN116433496A (en) | Image denoising method, device and storage medium | |
| Xue | Blind image deblurring: a review | |
| CN111416937B (en) | Image processing method, image processing device, storage medium and mobile equipment | |
| CN113344822B (en) | Image denoising method, device, terminal and storage medium | |
| CN113469908B (en) | Image noise reduction method, device, terminal and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |