CN110009563A - Image processing method and device, electronic equipment and storage medium - Google Patents
Image processing method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN110009563A CN110009563A CN201910239102.1A CN201910239102A CN110009563A CN 110009563 A CN110009563 A CN 110009563A CN 201910239102 A CN201910239102 A CN 201910239102A CN 110009563 A CN110009563 A CN 110009563A
- Authority
- CN
- China
- Prior art keywords
- image
- enhancing
- weight matrix
- matrix
- original
- 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 12
- 239000011159 matrix material Substances 0.000 claims abstract description 174
- 230000002708 enhancing effect Effects 0.000 claims abstract description 104
- 238000000034 method Methods 0.000 claims abstract description 73
- 238000012545 processing Methods 0.000 claims abstract description 53
- 238000013528 artificial neural network Methods 0.000 claims description 42
- 238000004590 computer program Methods 0.000 claims description 11
- 230000001965 increasing effect Effects 0.000 claims description 5
- 210000004218 nerve net Anatomy 0.000 claims description 2
- 239000004744 fabric Substances 0.000 claims 1
- 230000003014 reinforcing effect Effects 0.000 abstract description 6
- 238000004422 calculation algorithm Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 239000000047 product Substances 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 235000013399 edible fruits Nutrition 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000000903 blocking effect Effects 0.000 description 2
- 239000003086 colorant Substances 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000012805 post-processing Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 239000013589 supplement Substances 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20172—Image enhancement details
Landscapes
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The embodiment of the present invention provides a kind of image processing method and device, electronic equipment and storage medium, wherein the described method includes: obtaining the first image to original image processing, the data volume of the first image is less than the original image；Enhancing processing is carried out to the first image, obtains the first enhancing image；The first weight matrix is obtained based on the first image and the first enhancing image；By the size adjusting of first weight matrix at the size same or similar in size with the original image, the second weight matrix is obtained；Target image is generated based on the original image and second weight matrix, to realize the enhancing to the original image.When handling image by using method provided by the invention, it can maximumlly retain the detailed information of original image, to improve the reinforcing effect of image while guaranteeing processing speed.
Description
Technical field
This application involves field of image processing, in particular to a kind of image processing method and device, electronic equipment and storage
Medium.
Background technique
In recent years, image enhancement on the mobile terminals such as mobile phone using more and more extensive, especially for photo
Beautification.With the development of neural network, the effect of image enhancement has leaping for matter, but the place of the chip due to mobile terminal
Reason ability is limited, and the operand of neural network is very big, causes the efficiency of image enhancement not high.
In the prior art, in order to guarantee processing real-time, original image is usually subjected to a certain proportion of scaling, then
The enhancing that directly thumbnail of generation is calculated by neural network, and is exported neural network by the methods of linear interpolation
Thumbnail readjust be original image size.However, the thumbnail of enhancing to be adjusted to the size of original image
A large amount of detailed information can be lost in the process, and therefore, this method causes image enhancement effects poor, and further results in user's body
Test difference.
Summary of the invention
The embodiment of the present invention is designed to provide a kind of image processing method and device, electronic equipment and storage medium,
The detailed information of original image can maximumlly be retained while guaranteeing processing speed, to improve the enhancing effect of image
Fruit, and further promote user experience.
In order to solve the above-mentioned technical problem, the embodiment of the present invention adopts the technical scheme that
First aspect present invention provides a kind of image processing method, comprising:
To original image processing, the first image is obtained, the data volume of the first image is less than the original image；
Enhancing processing is carried out to the first image, obtains the first enhancing image；
The first weight matrix is obtained based on the first image and the first enhancing image；
By the size adjusting of first weight matrix at the size same or similar in size with the original image, obtain
To the second weight matrix；
Target image is generated based on the original image and second weight matrix, to realize to the original image
Enhancing.
Preferably, carrying out enhancing processing to the first image using neural network, the first enhancing image is obtained.
Preferably, it is described based on the first image and it is described first enhancing image obtain the first weight matrix, comprising:
The image array of the first image and the image array of the first enhancing image are obtained respectively；
The image array of image array and the first enhancing image to the first image a little obtain except operation
First weight matrix.
Preferably, before generating the target image, the method also includes:
Second weight matrix is filtered, so that the weight distribution of second weight matrix is more flat
It is sliding.
Second aspect of the present invention provides a kind of image processing apparatus, comprising:
Scaling processing module is configured to obtain the first image, the data volume of the first image is small to original image processing
In the original image；
Image enhancement module is configured to carry out enhancing processing to the first image, obtains the first enhancing image；
Weight matrix obtains module, is configured to the first image and the first enhancing image obtains the first weight
Matrix；
By the size adjusting of first weight matrix at the size same or similar in size with the original image, obtain
To the second weight matrix；
Target image generation module, is configured to the original image and second weight matrix generates target figure
Picture, to realize the enhancing to the original image.
Preferably, described image enhancing module is further configured to, the first image is carried out using neural network
Enhancing processing, obtains the first enhancing image.
It is further configured to preferably, the weight matrix obtains module, obtains the image of the first image respectively
The image array of matrix and the first enhancing image；
The image array of image array and the first enhancing image to the first image a little obtain except operation
First weight matrix.
Preferably, described device further includes filter module, it is configured to, before generating the target image, to institute
It states the second weight matrix to be filtered, so that the weight distribution of second weight matrix is more smooth.
Third aspect present invention provides a kind of electronic equipment, including memory, processor, is stored with meter on the memory
Calculation machine program, the processor realize above-mentioned method when executing the computer program on the memory.
Fourth aspect present invention provides a kind of storage medium, is stored with computer program, and the computer program is processed
Device realizes above-mentioned method when executing.
The beneficial effect of the embodiment of the present invention is:
Image processing method provided by the invention can maximumlly retain original graph while guaranteeing processing speed
The detailed information of picture to improve the reinforcing effect of image, and further promotes user experience.
Detailed description of the invention
Fig. 1 is the flow chart of the image processing method of the embodiment of the present invention；
Fig. 2 is the logic diagram of the image processing apparatus of the embodiment of the present invention.
Specific embodiment
Various schemes and feature of the invention are described herein with reference to attached drawing.
It should be understood that various modifications can be made to the embodiment invented herein.Therefore, description above should not regard
To limit, and only as the example of embodiment.Those skilled in the art will expect within the scope and spirit of this invention
Other modifications.
The attached drawing being included in the description and forms part of the description shows the embodiment of the present invention, and with it is upper
What face provided is used to explain the present invention substantially description and the detailed description given below to embodiment of the invention together
Principle.
It is of the invention by the description of the preferred form with reference to the accompanying drawings to the embodiment for being given as non-limiting example
These and other characteristic will become apparent.
Although being also understood that invention has been described referring to some specific examples, those skilled in the art
Member realizes many other equivalents of the invention in which can determine, they have feature as claimed in claim and therefore all
In the protection scope defined by whereby.
When read in conjunction with the accompanying drawings, in view of following detailed description, above and other aspect of the invention, feature and advantage will become
It is more readily apparent.
Specific embodiments of the present invention are described hereinafter with reference to attached drawing；It will be appreciated, however, that the embodiment invented is only
Various ways implementation can be used in example of the invention.Known and/or duplicate function and structure and be not described in detail to avoid
Unnecessary or extra details makes the present invention smudgy.Therefore, the specific structural and functionality invented herein is thin
Section is not intended to restrictions, but as just the basis of claim and representative basis be used to instructing those skilled in the art with
Substantially any appropriate detailed construction diversely uses the present invention.
This specification can be used phrase " in one embodiment ", " in another embodiment ", " in another embodiment
In " or " in other embodiments ", it can be referred to one or more of identical or different embodiment according to the present invention.
As shown in Figure 1, first embodiment of the invention provides a kind of image processing method, comprising:
To original image processing, the first image is obtained, the data volume of the first image is less than the original image；
Enhancing processing is carried out to the first image, obtains the first enhancing image；
The first weight matrix is obtained based on the first image and the first enhancing image；
By the size adjusting of first weight matrix at the size same or similar in size with the original image, obtain
To the second weight matrix；
Target image is generated based on the original image and second weight matrix, to realize to the original image
Enhancing.
In the above-described embodiments, by being handled original image to obtain the first image, thus by data volume it is small
Object of one image as image enhancement processing, the speed that enhancing can be made to handle in this way become faster, and obtain after being then based on enhancing
To the first enhancing image and the first image obtain the first weight matrix, then by the size of the first weight matrix be tuned into it is original
Image is same or similar, obtains the second weight matrix, finally generates target figure based on original image and second weight matrix
Picture.The method applied in the present invention, there is no the methods of linear interpolation used in background technique thus directly to enhanced
The size that image is readjusted as original image；But by first obtaining the first weight matrix, then pass through the first weight matrix
The second weight matrix is obtained, generates target image finally by original image and second weight matrix, through the invention
This method target image generated will not lose a large amount of detailed information, can maximumlly retain the thin of original image
Information is saved, to improve the reinforcing effect of image, and further promotes user experience.
In the present invention, the image that can be absorbed to any electronic equipment is handled, and is not appointed to the electronic equipment
What is limited, and is only illustratively to be illustrated so that electronic equipment is mobile phone as an example below.
It the use of mobile phone shooting photo has been public demand since present mobile phone is more more and more universal.But since mobile phone is set
Standby is limited, ability of taking pictures much deficiency slr cameras, so in order to shoot good-looking photo with mobile phone, it is usually used
Algorithm carries out post-processing to the photo that mobile phone is shot, and such treatment process is called image enhancement.Institute in above-described embodiment
The original image of the original image stated here refers to the digital picture directly shot by mobile phone.Digital picture is by a series of
The matrix of number composition, the color of each value representative image respective pixel position in this matrix.We are to original original image
It is actually some operations carried out to this matrix that image, which carries out enhancing operation, and the result after operation is exactly certain in matrix
Value is changed.So the form of expression of a width digital picture in a computer should be the form as shown in formula (I).
Wherein Ii,jRefer to the color of the i-th row jth column respective pixel position in image (if it is black white image, here
It can be understood as the degree of black and white, if we describe the degree of black and white with 255 grades, then I=255 just represents the position
Setting is white, and it is black that I=0, which just represents this position, and color image is similar), the value range of i and j are exactly picture size, such as
As soon as fruit is the image that resolution ratio is 640*480, then what this matrix should be made of the number that 640 rows 480 arrange.
In one embodiment of the invention, to original image processing, the first image, the data of the first image are obtained
Amount is less than the original image；It specifically refers to, by the matrixing of the original image matrix smaller than the matrix at a size.
Enhancing processing is carried out to it to facilitate, by handling relatively small first image of data volume, can be made compared to directly right
The speed that original image carries out enhancing processing becomes faster, and has saved the time.
Wherein, the first image is construed as the original image thumbnail of original image, and the data volume of the first image is either
The data volume of original image is construed as the resolution ratio of image；Original image is handled to obtain the mistake of the first image
Miscellaneous picture, is actually uniformly scaled to the process of the picture of a fixed size by journey, such as mobile phone
Various photos can be stored in photograph album, their resolution ratio is all different, some photos are 640*480, some photos
It is 800*600, some photos are 1280*800, then need these photos being uniformly scaled to 500*500.Here 500*
500 photo is exactly above-mentioned the first image obtained by original image.Then enhancing processing is carried out to the first image again, obtained
Enhanced processing result.Enhancing treated output result be also a character matrix.
Image enhancement is actually the content passed through in the character matrix for changing image, to change respective pixel in image
The color of position, to achieve the effect that enhancing.For example, the result of an image enhancement operation may be it is following in this way: where
I indicates that the matrix of original image, I ' indicate the matrix obtained after original image enhancing；
It can be seen that some numerical value in the matrix of enhanced image are changed, and these variations are set by electronics
It is exactly that the colors of certain positions in image is changed that the display of standby display device, which is embodied on image,.Currently a popular is each
The U.S. face method of kind is exactly one kind of image enhancement, and the color of application on human skin position is exactly become brighter.Our applied field
Scape is focused primarily on mobile phone terminal, so U.S. face algorithm just simply can also be understood as algorithm for image enhancement.It continues the above
Example, image enchancing method are exactly that some values in the matrix original image are changed, and this variation has foundation, root
According to the regular also difference of its transformation of different image enchancing methods.Comparing traditional certain methods is exactly simply to increase whole figure
The color saturation (color saturation can be calculated by image digitization matrix) of picture, or do histogram equalization,
Image gamma transformation etc..Currently, carrying out image enhancement using neural network is also popular method, but no matter what uses
Kind method carries out enhancing processing to image, and common practice is directly to export final enhancing figure, then this is increased in the prior art
Strong figure reverts to original size and has just obtained final result.But this mode is thin due to that can bring during reduction
Section loss, although the reinforcing effect that neural network obtains is fine, enhanced image has lost after re-scaling
Part details causes final result undesirable.
For example the enhanced matrix of a 2*2 (matrix I) is reduced directly the square to original 4*4 (matrix I ')
Battle array, it would be possible that form be following such:
Be scaled to one big figure if it is by small figure, then the value for certainly existing some pixels be it is unknown, this is just needed
What very important person was supplements these values, and rule used in its supplementary unknown of different image-scaling methods is different.It is most simple
Single method is a kind of method for being called neighbor interpolation, and this method is simply one of the value adjacent position of unknown position
Value substitutes, such as above example, if with neighbor interpolation method, obtained result will be it is following in this way:
As soon as thereby realizing a matrix being exaggerated by 2*2 to 4*4, it is so-called original in conventional method also to have obtained
The enhanced image of image.There are also bilinear interpolations, cubic etc. for other some Zoom methods.But whether which
Kind interpolation method cannot all reflect image due detailed information originally completely, can only allow these as far as possible
The information artificially filled up more meets really, so from image, amplified image it is possible that mosaic or
Situations such as person deforms.This phenomenon is just called bring loss of detail during scaling.Therefore, used straight in the prior art
The final enhancing figure of output is connect, then this enhancing figure is reverted into original size to obtain the method for final result and can exist seriously
Loss of detail.
The method applied in the present invention, which has been evaded, directly zooms in and out enhancing figure, but the first weight is obtained by calculation
Then first weight matrix is zoomed in and out the method for obtaining the second weight matrix by matrix.Although obtained after scaling the
Two weight matrix also bring along certain loss, but by by after the information of the second weight matrix combination original image, it can
To allow the loss of detail of result to be reduced to minimum.
In another embodiment of the present invention, enhancing processing is carried out to the first image using neural network, obtained
First enhancing image.
First image is sent into neural network, the character matrix of the first image is grasped by a series of operations of neural network
Make, obtains the output of neural network as a result, obtaining the first enhancing image.
The method of neural network is exactly currently a popular deep learning method, that is, the common method of artificial intelligence it
One.It simply can be understood as this and set formed operated by a series of addition subtraction multiplication and division, when a matrix is sent into neural network
Later, neural network will carry out a series of addition subtraction multiplication and division operation to each value in matrix, and some of them operation may also can
Change the size of matrix.In order to allow neural network that can handle miscellaneous data, need to guarantee to be sent into neural network every time
The size of matrix be it is unified, in this way could the value to corresponding position in matrix carry out corresponding operation.So in simple terms, it is defeated
The size for entering the matrix of neural network is preferably maintained in unified, and the present invention does not have any limit to the size of matrix of input neural network
System, as long as it is smaller than the data volume of original image, meet the standard that neural network is able to carry out processing, for example all can only be
The matrix of 640*480 size, that is to say, that the matrix size of the first image is 640*480.Same reason, neural network are passed through
It is a series of be uniformly processed after, the content of output is also unified, for example neural network may be a series of 640*480
Matrix disposal after all become the matrixes of 1080*720.So the size of in general neural network output and input is all
It is unified.
In other embodiments of the invention, it is described based on the first image and it is described first enhancing image obtain first
Weight matrix, comprising:
The image array of the first image and the image array of the first enhancing image are obtained respectively；
The image array of image array and the first enhancing image to the first image a little obtain except operation
First weight matrix.
In the above-described embodiments, the calculation of the first weight matrix M can there are many kinds of, the present invention does not make this specifically
It limits, is exemplarily only calculated in such a way that matrix dot removes, to obtain the first weight matrix.
Wherein i, j respectively indicate the value of the i-th row jth column corresponding position of corresponding matrix.O indicates the first enhancing figure
Picture, I indicate the first image.Citing:
Then
Here the first weight matrix indicates a difference degree of the first image and the first image after enhancing, intuitively comes
See to be exactly a matrix, in this matrix each value correspond to original image to enhancing figure corresponding position mapping relations.
In another embodiment of the present invention, before generating the target image, the method also includes:
Second weight matrix is filtered, so that the weight distribution of second weight matrix is more flat
It is sliding.
In the present invention, for by the size adjusting of first weight matrix at identical as the size of the original image or
Method used in similar size does not have any restrictions, as long as can by the size adjusting of the first weight matrix at the original
The size of beginning image is same or similar, such as can be linear interpolation method or neighbor interpolation method etc..But
It can carry out Gaussian smoothing again after carrying out interpolation in embodiments of the present invention, to make up interpolation method bring loss of detail,
Even mosaic effect brought by interpolation method, (as blocking artifact, all pixels point is all same face in certain fritter
Color, so seeming is exactly that a fritter is the same), the more smooth of value transition between block and block is allowed, to allow the second weight square
The weight distribution of battle array is more smooth, and the image seemed in this way is more smooth.
In one embodiment of the invention, by the size adjusting of first weight matrix at the original image
Size same or similar in size obtains the second weight matrix, comprising:
Obtain the image array of the original image；
The size of first weight matrix is enlarged into or phase identical as the size of the image array of the original image
Closely, to obtain second weight matrix.
In embodiment provided by the present invention, since the size of the image exported by Processing with Neural Network may be
One fixed dimension, i.e., the first enhancing image may be a fixed dimension, but the fixed dimension may not be with the image of original image
Size is identical, so needing for the size of first weight matrix to be enlarged into the size with the image array of the original image
It is same or similar, to obtain second weight matrix.Then again based on the original image and second weight matrix into
The dot product of row corresponding position is operated to generate target image, to realize the enhancing to the original image.
It is described raw based on the original image and second weight matrix in other embodiments provided by the present invention
At target image, to realize the enhancing to the original image, comprising:
Image array and second weight matrix to the original image carry out dot product operation, obtain the target figure
The image array of picture；
Image array based on the target image generates the target image, to realize the increasing to the original image
By force.
Based on the same inventive concept, as shown in Fig. 2, second embodiment of the invention provides a kind of image processing apparatus, packet
It includes:
Scaling processing module is configured to obtain the first image, the data volume of the first image is small to original image processing
In the original image；
Image enhancement module is coupled with the scaling processing module, is configured to carry out at enhancing the first image
Reason, obtains the first enhancing image；
Weight matrix obtains module, couples, configures with the scaling processing module and described image enhancing module respectively
To obtain the first weight matrix based on the first image and the first enhancing image；
By the size adjusting of first weight matrix at the size same or similar in size with the original image, obtain
To the second weight matrix；
Target image generation module obtains module with the weight matrix and couples, is configured to the original image
Target image is generated with second weight matrix, to realize the enhancing to the original image.
In the above-described embodiments, by being handled original image to obtain the first image, thus by data volume it is small
Object of one image as image enhancement processing, the speed that enhancing can be made to handle in this way become faster, and obtain after being then based on enhancing
To the first enhancing image and the first image obtain the first weight matrix, then by the size of the first weight matrix be tuned into it is original
Image is same or similar, obtains the second weight matrix, finally generates target figure based on original image and second weight matrix
Picture.Device of the present invention, there is no the sizes directly readjusted to enhanced image as original image；But it is logical
After first obtaining the first weight matrix, the second weight matrix is then obtained by the first weight matrix, finally by original image and
Second weight matrix generates target image, and this device target image generated through the invention will not lose largely
Detailed information, can maximumlly retain the detailed information of original image, to improve the reinforcing effect of image, go forward side by side one
Step promotes user experience.
In the present invention, the image that can be absorbed to any electronic equipment is handled, and is not appointed to the electronic equipment
What is limited, and is only illustratively to be illustrated so that electronic equipment is mobile phone as an example below.
It the use of mobile phone shooting photo has been public demand since present mobile phone is more more and more universal.But since mobile phone is set
Standby is limited, ability of taking pictures much deficiency slr cameras, so in order to shoot good-looking photo with mobile phone, it is usually used
Algorithm carries out post-processing to the photo that mobile phone is shot, and such treatment process is called image enhancement.Institute in above-described embodiment
The original image of the original image stated here refers to the digital picture directly shot by mobile phone.Digital picture is by a series of
The matrix of number composition, the color of each value representative image respective pixel position in this matrix.We are to original original image
It is actually some operations carried out to this matrix that image, which carries out enhancing operation, and the result after operation is exactly certain in matrix
Value is changed.So the form of expression of a width digital picture in a computer should be the form as shown in formula (I).
Wherein Ii,jRefer to the color of the i-th row jth column respective pixel position in image (if it is black white image, here
It can be understood as the degree of black and white, if we describe the degree of black and white with 255 grades, then I=255 just represents the position
Setting is white, and it is black that I=0, which just represents this position, and color image is similar), the value range of i and j are exactly picture size, such as
As soon as fruit is the image that resolution ratio is 640*480, then what this matrix should be made of the number that 640 rows 480 arrange.
In one embodiment of the invention, to original image processing, the first image, the data of the first image are obtained
Amount is less than the original image；It specifically refers to, by the matrixing of the original image matrix smaller than the matrix at a size.
Enhancing processing is carried out to it to facilitate, by handling relatively small first image of data volume, can be made compared to directly right
The speed that original image carries out enhancing processing becomes faster, and has saved the time.
Wherein, the first image is construed as the original image thumbnail of original image, and the data volume of the first image is either
The data volume of original image is construed as the resolution ratio of image；Original image is handled to obtain the mistake of the first image
Miscellaneous picture, is actually uniformly scaled to the process of the picture of a fixed size by journey, such as mobile phone
Various photos can be stored in photograph album, their resolution ratio is all different, some photos are 640*480, some photos
It is 800*600, some photos are 1280*800, then need these photos being uniformly scaled to 500*500.Here 500*
500 photo is exactly above-mentioned the first image obtained by original image.Then enhancing processing is carried out to the first image again, obtained
Enhanced processing result.Enhancing treated output result be also a character matrix.
Image enhancement is actually the content passed through in the character matrix for changing image, to change respective pixel in image
The color of position, to achieve the effect that enhancing.For example, the result of an image enhancement operation may be it is following in this way: where
I indicates that the matrix of original image, I ' indicate the matrix obtained after original image enhancing；
It can be seen that some numerical value in the matrix of enhanced image are changed, and these variations are set by electronics
It is exactly that the colors of certain positions in image is changed that the display of standby display device, which is embodied on image,.Currently a popular is each
The U.S. face method of kind is exactly one kind of image enhancement, and the color of application on human skin position is exactly become brighter.Our applied field
Scape is focused primarily on mobile phone terminal, so U.S. face algorithm just simply can also be understood as algorithm for image enhancement.It continues the above
Example, image enchancing method are exactly that some values in the matrix original image are changed, and this variation has foundation, root
According to the regular also difference of its transformation of different image enchancing methods.Comparing traditional certain methods is exactly simply to increase whole figure
The color saturation (color saturation can be calculated by image digitization matrix) of picture, or do histogram equalization,
Image gamma transformation etc..Currently, carrying out image enhancement using neural network is also popular method, but no matter what uses
Kind method carries out enhancing processing to image, and common practice is directly to export final enhancing figure, then this is increased in the prior art
Strong figure reverts to original size and has just obtained final result.But this mode is thin due to that can bring during reduction
Section loss, although the reinforcing effect that neural network obtains is fine, enhanced image has lost after re-scaling
Part details causes final result undesirable.
For example the enhanced matrix of a 2*2 (matrix I) is reduced directly the square to original 4*4 (matrix I ')
Battle array, it would be possible that form be following such:
Be scaled to one big figure if it is by small figure, then the value for certainly existing some pixels be it is unknown, this is just needed
What very important person was supplements these values, and rule used in its supplementary unknown of different image-scaling methods is different.It is most simple
Single method is a kind of method for being called neighbor interpolation, and this method is simply one of the value adjacent position of unknown position
Value substitutes, such as above example, if with neighbor interpolation method, obtained result will be it is following in this way:
As soon as thereby realizing a matrix being exaggerated by 2*2 to 4*4, it is so-called original in conventional method also to have obtained
The enhanced image of image.There are also bilinear interpolations, cubic etc. for other some Zoom methods.But whether which
Kind interpolation method cannot all reflect image due detailed information originally completely, can only allow these as far as possible
The information artificially filled up more meets really, so from image, amplified image it is possible that mosaic or
Situations such as person deforms.This phenomenon is just called bring loss of detail during scaling.Therefore, used straight in the prior art
The final enhancing figure of output is connect, then this enhancing figure is reverted into original size to obtain the method for final result and can exist seriously
Loss of detail.
Device of the present invention, which has been evaded, directly zooms in and out enhancing figure, but the first weight is obtained by calculation
Then first weight matrix is zoomed in and out the method for obtaining the second weight matrix by matrix.Although obtained after scaling the
Two weight matrix also bring along certain loss, but by by after the information of the second weight matrix combination original image, it can
To allow the loss of detail of result to be reduced to minimum.
In another embodiment provided by the invention, described image enhancing module is further configured to, using nerve net
Network carries out enhancing processing to the first image, obtains the first enhancing image.
First image is sent into neural network, the character matrix of the first image is grasped by a series of operations of neural network
Make, obtains the output of neural network as a result, obtaining the first enhancing image.
The method of neural network is exactly currently a popular deep learning method, that is, the common method of artificial intelligence it
One.It simply can be understood as this and set formed operated by a series of addition subtraction multiplication and division, when a matrix is sent into neural network
Later, neural network will carry out a series of addition subtraction multiplication and division operation to each value in matrix, and some of them operation may also can
Change the size of matrix.In order to allow neural network that can handle miscellaneous data, need to guarantee to be sent into neural network every time
The size of matrix be it is unified, in this way could the value to corresponding position in matrix carry out corresponding operation.So in simple terms, it is defeated
The size for entering the matrix of neural network is preferably maintained in unified, and the present invention does not have any limit to the size of matrix of input neural network
System, as long as it is smaller than the data volume of original image, meet the standard that neural network is able to carry out processing, for example all can only be
The matrix of 640*480 size, that is to say, that the matrix size of the first image is 640*480.Same reason, neural network are passed through
It is a series of be uniformly processed after, the content of output is also unified, for example neural network may be a series of 640*480
Matrix disposal after all become the matrixes of 1080*720.So the size of in general neural network output and input is all
It is unified.
In other embodiments of the invention, the weight matrix obtains module and is further configured to, respectively described in acquisition
The image array of the image array of first image and the first enhancing image；
The image array of image array and the first enhancing image to the first image a little obtain except operation
First weight matrix.
In the above-described embodiments, the calculation of the first weight matrix M can there are many kinds of, the present invention does not make this specifically
It limits, is exemplarily only calculated in such a way that matrix dot removes, to obtain the first weight matrix.
Wherein i, j respectively indicate the value of the i-th row jth column corresponding position of corresponding matrix.O indicates the first enhancing figure
Picture, I indicate the first image.Citing:
Then
Here the first weight matrix indicates a difference degree of the first image and the first image after enhancing, intuitively comes
See to be exactly a matrix, in this matrix each value correspond to original image to enhancing figure corresponding position mapping relations.
In another embodiment of the present invention, described device further includes filter module, is configured to, and the mesh is being generated
Before logo image, second weight matrix is filtered, so that the weight distribution of second weight matrix is more
Smoothly.
In the present invention, for by the size adjusting of first weight matrix at identical as the size of the original image or
Method used in similar size does not have any restrictions, as long as can by the size adjusting of the first weight matrix at the original
The size of beginning image is same or similar, such as can be linear interpolation method or neighbor interpolation method etc..But
It can carry out Gaussian smoothing again after carrying out interpolation in embodiments of the present invention, to make up interpolation method bring loss of detail,
Even mosaic effect brought by interpolation method, (as blocking artifact, all pixels point is all same face in certain fritter
Color, so seeming is exactly that a fritter is the same), the more smooth of value transition between block and block is allowed, to allow the second weight square
The weight distribution of battle array is more smooth, and the image seemed in this way is more smooth.
In one embodiment of the invention, the weight matrix obtains module and is additionally configured to:
Obtain the image array of the original image；
The size of first weight matrix is enlarged into or phase identical as the size of the image array of the original image
Closely, to obtain second weight matrix.
In embodiment provided by the present invention, since the size of the image exported by Processing with Neural Network may be
One fixed dimension, i.e., the first enhancing image may be a fixed dimension, but the fixed dimension may not be with the image of original image
Size is identical, so needing for the size of first weight matrix to be enlarged into the size with the image array of the original image
It is same or similar, to obtain second weight matrix.Then again based on the original image and second weight matrix into
The dot product of row corresponding position is operated to generate target image, to realize the enhancing to the original image.
In other embodiments provided by the present invention, the target image generation module is additionally configured to:
Image array and second weight matrix to the original image carry out dot product operation, obtain the target figure
The image array of picture；
Image array based on the target image generates the target image, to realize the increasing to the original image
By force.
Based on inventive concept identical with above-mentioned image processing method, third embodiment of the invention provides a kind of electronics and sets
It is standby, including memory, processor, computer program is stored on the memory, the processor is executing the memory
On computer program when realize above-mentioned method.
Based on inventive concept identical with above-mentioned image processing method, four embodiment of the invention provides a kind of storage Jie
Matter, is stored with computer program, and the computer program realizes above-mentioned method when being executed by processor.
Above embodiments are only exemplary embodiment of the present invention, are not used in the limitation present invention, protection scope of the present invention
It is defined by the claims.Those skilled in the art can within the spirit and scope of the present invention make respectively the present invention
Kind modification or equivalent replacement, this modification or equivalent replacement also should be regarded as being within the scope of the present invention.
Claims (10)
1. a kind of image processing method, comprising:
To original image processing, the first image is obtained, the data volume of the first image is less than the original image；
Enhancing processing is carried out to the first image, obtains the first enhancing image；
The first weight matrix is obtained based on the first image and the first enhancing image；
By the size adjusting of first weight matrix at the size same or similar in size with the original image, is obtained
Two weight matrix；
Target image is generated based on the original image and second weight matrix, to realize the increasing to the original image
By force.
2. the method according to claim 1, wherein being carried out at enhancing using neural network to the first image
Reason, obtains the first enhancing image.
3. the method according to claim 1, wherein described schemed based on the first image with first enhancing
As obtaining the first weight matrix, comprising:
The image array of the first image and the image array of the first enhancing image are obtained respectively；
The image array of image array and the first enhancing image to the first image a little obtain described except operation
First weight matrix.
4. the method according to claim 1, wherein the method is also wrapped before generating the target image
It includes:
Second weight matrix is filtered, so that the weight distribution of second weight matrix is more smooth.
5. a kind of image processing apparatus, comprising:
Scaling processing module is configured to obtain the first image to original image processing, and the data volume of the first image is less than institute
State original image；
Image enhancement module is configured to carry out enhancing processing to the first image, obtains the first enhancing image；
Weight matrix obtains module, is configured to the first image and the first enhancing image obtains the first weight square
Battle array；
By the size adjusting of first weight matrix at the size same or similar in size with the original image, is obtained
Two weight matrix；
Target image generation module, is configured to the original image and second weight matrix generates target image, with
Realize the enhancing to the original image.
6. image processing apparatus according to claim 5, described image enhancing module is further configured to, using nerve net
Network carries out enhancing processing to the first image, obtains the first enhancing image.
7. image processing apparatus according to claim 5, the weight matrix obtains module and is further configured to, and obtains respectively
Take the image array of the first image and the image array of the first enhancing image；
The image array of image array and the first enhancing image to the first image a little obtain described except operation
First weight matrix.
8. image processing apparatus according to claim 5, described device further includes filter module, is configured to, and is being generated
Before the target image, second weight matrix is filtered, so that the weight of second weight matrix point
Cloth is more smooth.
9. a kind of electronic equipment, including memory, processor, computer program is stored on the memory, which is characterized in that
The processor is realized according to any one of claims 1 to 4 when executing the computer program on the memory
Method.
10. a kind of storage medium, is stored with computer program, which is characterized in that when the computer program is executed by processor
Realize method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910239102.1A CN110009563B (en) | 2019-03-27 | 2019-03-27 | Image processing method and device, electronic device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910239102.1A CN110009563B (en) | 2019-03-27 | 2019-03-27 | Image processing method and device, electronic device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110009563A true CN110009563A (en) | 2019-07-12 |
| CN110009563B CN110009563B (en) | 2021-02-19 |
Family
ID=67168517
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910239102.1A Active CN110009563B (en) | 2019-03-27 | 2019-03-27 | Image processing method and device, electronic device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110009563B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022027197A1 (en) * | 2020-08-03 | 2022-02-10 | SZ DJI Technology Co., Ltd. | Systems and methods for processing image |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060232823A1 (en) * | 2005-04-13 | 2006-10-19 | Hooper David S | Image contrast enhancement |
| CN102622732A (en) * | 2012-03-14 | 2012-08-01 | 上海大学 | Front-scan sonar image splicing method |
| CN103345732A (en) * | 2013-07-26 | 2013-10-09 | 电子科技大学 | Pulse coupled neural network (PCNN) image enhancement algorithm and device based on Contourlet transformation |
-
2019
- 2019-03-27 CN CN201910239102.1A patent/CN110009563B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060232823A1 (en) * | 2005-04-13 | 2006-10-19 | Hooper David S | Image contrast enhancement |
| CN102622732A (en) * | 2012-03-14 | 2012-08-01 | 上海大学 | Front-scan sonar image splicing method |
| CN103345732A (en) * | 2013-07-26 | 2013-10-09 | 电子科技大学 | Pulse coupled neural network (PCNN) image enhancement algorithm and device based on Contourlet transformation |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022027197A1 (en) * | 2020-08-03 | 2022-02-10 | SZ DJI Technology Co., Ltd. | Systems and methods for processing image |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110009563B (en) | 2021-02-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Klatzer et al. | Learning joint demosaicing and denoising based on sequential energy minimization | |
| Delbracio et al. | Mobile computational photography: A tour | |
| CN108604369B (en) | Method, device and equipment for removing image noise and convolutional neural network | |
| CN104376542B (en) | A kind of image enchancing method | |
| Ren et al. | Single image super-resolution via adaptive high-dimensional non-local total variation and adaptive geometric feature | |
| CN109255769A (en) | The training method and training pattern and image enchancing method of image enhancement network | |
| US7983511B1 (en) | Methods and apparatus for noise reduction in digital images | |
| CN108765343A (en) | Method, apparatus, terminal and the computer readable storage medium of image procossing | |
| CN109544482A (en) | A kind of convolutional neural networks model generating method and image enchancing method | |
| CN110059815B (en) | Artificial intelligence reasoning computing equipment | |
| KR102606824B1 (en) | Apparatus comprising multi-camera and image processing method thereof | |
| KR102095443B1 (en) | Method and Apparatus for Enhancing Image using Structural Tensor Based on Deep Learning | |
| RU2697928C1 (en) | Superresolution of an image imitating high detail based on an optical system, performed on a mobile device having limited resources, and a mobile device which implements | |
| CN105389776B (en) | Image scaling techniques | |
| CN113822830A (en) | Multi-exposure image fusion method based on depth perception enhancement | |
| Nam et al. | Modelling the scene dependent imaging in cameras with a deep neural network | |
| CN106162131B (en) | A kind of real time image processing | |
| CN113781318A (en) | Image color mapping method and device, terminal equipment and storage medium | |
| CN109118436A (en) | Picture tone adaptation method and corresponding electronic equipment and storage medium | |
| Hsieh et al. | Variational contrast-saturation enhancement model for effective single image dehazing | |
| CN110009563A (en) | Image processing method and device, electronic equipment and storage medium | |
| CN111951171A (en) | HDR image generation method and device, readable storage medium and terminal equipment | |
| CN105590294B (en) | A kind of image processing method and electronic equipment | |
| Du et al. | Low-light image enhancement and denoising via dual-constrained Retinex model | |
| US9619864B2 (en) | Image processing apparatus and method for increasing sharpness of images |
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 |