CN109710791A - A kind of multi-source color image color moving method based on significant filter - Google Patents
A kind of multi-source color image color moving method based on significant filter Download PDFInfo
- Publication number
- CN109710791A CN109710791A CN201811535193.5A CN201811535193A CN109710791A CN 109710791 A CN109710791 A CN 109710791A CN 201811535193 A CN201811535193 A CN 201811535193A CN 109710791 A CN109710791 A CN 109710791A
- Authority
- CN
- China
- Prior art keywords
- image
- color
- region
- color image
- channel
- 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.)
- Pending
Links
Abstract
The invention discloses a kind of multi-source color image color moving method based on significant filter, belongs to digital image processing field.Color transfer method proposed by the present invention solves single width source images and provides the limited defect of color, with reference to specified color image, carries out color adjustment to current color image, result images are showed with tone similar with color image so that treated.The color transfer between color image can be preferably realized by this method, and the execution time is shorter, does not need manually to be labeled image, can obtain preferable color transfer effect.Compared with traditional color transfer method, though the result that context of methods generates depends on color image, and it is insensitive to distribution of color information.
Description
Technical field
The present invention relates to digital image processing fields, and in particular to a kind of multi-source color image face based on significant filter
Colour migration shifting method.
Background technique
The fast development of modern society causes people to be surrounded already by miscellaneous information, and human eye passes through to image
Observation, directly or indirectly obtains a large amount of visual information.Digital image processing techniques are important in field of image processing
Component part is a key areas of computer technology research.Digital image processing techniques are very multi-disciplinary bases, are had
Very strong practical value is widely used in the various fields such as medicine, traffic, medium, education.In image procossing one most often
The problem of seeing is the color for changing image, is allowed to become our required colors.
From the point of view of the viewing angle of human eye, color is one of the essential characteristic of human perception and differentiation different objects, generally
For, identical object has same or similar shade and distribution, and different objects passes through shade and distribution
Difference so that the mankind is perceived to being distinguished.And color characteristic has as one of most important feature of object
Certain stability, it is all insensitive to the dimensional variation of image, rotation, translation, and indicate and calculate all fairly simple.Color
Migration is to merge the shape information of the colouring information of image A and image B, generates another piece image C, and makes the knot generated
Fruit image not only active image A colouring information, but also have the shape information of target image B, it is allowed to seem the view for being more in line with the mankind
Feel esthetic requirement.Wherein image A is known as color image, is also source images, and image B is known as shape image, is also target image.Face
Color is one of most important visual information, and color transfer is an important topic of Digital Image Processing, by changing target figure
The colouring information of picture makes target image that specific variation occur in terms of color, brightness and art, changes target image
Color style, prominent image subject enhances image appeal.
However there are the target image of complex scene, the color that the single source images to match provide is limited.In addition, existing
Algorithm be not the corresponding region that the color of source images is transferred to target image in a manner of a kind of perception alignment.Set forth herein
A kind of method of multi-source colour migration between color image based on significant filter.Using Scale invariant features transform
(Scale-Invariant Feature Transform, SIFT) algorithm carries out image retrieval in image library, and it is suitable to find
Two width source images.Then, input picture is divided into foreground area and background area by the notable figure obtained by significant filter
Domain.Finally, finding best region matching pair using weighted color migration algorithm, reaches the multi-source color based on significant filter and move
The effect of shifting.
Summary of the invention
The technical scheme is that a kind of multi-source color image color moving method based on significant filter, described
Method includes the following steps:
S1. characteristics of image is first extracted, then carries out the matching between feature, is obtained and the higher two width shape of shape image similarity
Shape image.SIFT feature extraction process is divided into two processes: first extracting feature offline and stores, then online quick-searching.
Detailed process is as follows: the former is first extracted the SIFT feature of image and stores into database, is later image retrieval procedure
Spare retrieval data are provided.After the completion of off-line procedure, then to target retrieval image carry out SIFT feature extraction, with before from
The database that line is completed carries out characteristic matching, completes on-line retrieval process.Finally, the matched points of statistical picture, matched number
Mesh, and to turn out the image being found in the database more similar to shape image.
S2. super-pixel segmentation is carried out to the obtained color image of step S1 and given shape image.By image from RGB face
For color space transformation to CIE-l α β color space, (l, α, the β) color value and (x, y) coordinate for corresponding to each pixel form one 5 dimension
The similitude of vector V [l, α, β, x, y], two pixels can be measured by their vector distance, and distance is bigger, and similitude is got over
It is small.
Algorithm firstly generates M seed point, and then the detection range seed point is most in the surrounding space of each seed point
Close several pixels, by they be classified as with the seed point one kind, all sort out until all pixels point and finish.Then this K are calculated
The average vector value of all pixels point, retrieves K cluster centre in super-pixel, then again with this K center removal search its
Surrounding and its most similar several pixel, all pixels retrieve K super-pixel after all having sorted out, update cluster centre,
Iteration again, repeatedly until convergence.
S3. brightness is carried out to the obtained color image of step S2 and shape image to remap.During Region Matching, source
The very big situation of the Luminance Distribution difference of image and each pixel of target image often causes large effect to matching result,
Therefore before matching, need first to implement brightness to the channel l of every color image to remap, formula is as follows:
In formulaWithFor the mean value and standard deviation in the channel l in source images s (k)；WithFor the channel target image l
Mean value and variance.For lkNew matching brightness value.
S4. the most matched region pair of color image and shape image is selected.Using significant filter by target image and source
Image segmentation is at foreground and background.The region of high saliency value is known as prospect, and the region of low saliency value is background.In addition, by setting
The segmentation threshold for setting input picture is twice of corresponding notable figure mean value, and picture breakdown Cheng Youqi average color is indicated tight
It gathers, perceives uniform element.By calculating color image region si(k), i=1, the mean value vector of 2 ..., MAnd shape
Image-region tj, the color mean value vector of j=1,2 ..., MBetween Euclidean distance be used as according to being judged, distance is most
Two small regions are best match region pair.The distance between the color mean value in color image and shape image region calculates
Formula is as follows:
By above formula, in that case it can be decided that most matched region is to (tj,si), in tjK candidate regions between si be similar source region
Domain is described as follows:
(tj,si)←min({f(tj,si(k)) | 0 < k≤K≤M })
By collecting siIn pixel, we can obtain a compound source images S.
S5. it by the shape image region of the color transfer in each color image region to Corresponding matching, specifically uses such as
Lower formula calculates the transformed value in the channel color image α β:
α in formulatAnd βtIt is target image α, the pixel in the channel β respectively；WithIt is the area target image jKuai respectively
The mean value in the channel the α in domain, β.WithIt is the α in i-th piece of region of color image, the mean value in the channel β respectively；WithPoint
It is not the α in target image jth block region, the standard variance in the channel β；WithIt is the α in i-th piece of region of color image respectively,
The standard variance in the channel β.
The conversion value in the channel color image l is specifically calculated using following formula:
Each Regional Color Transfer result images are synthesized into complete image, then the image of synthesis is converted from the space α β l
Return RGB color.
Compared with prior art, the present invention has the advantage that and the utility model has the advantages that
1, the new method that a kind of pair of color image carries out color transfer is devised, this method treatment effect is preferable, realizes speed
Degree is very fast；
2, it is realized using Euclidean distance and is mapped between two images, finally use the linear weighted function group of mean value and standard variance
The method of conjunction completes color transfer synthesis.
Though 3, the result that context of methods generates depends on color image, Ke Yitong insensitive to distribution of color information
The proportionality coefficient of adjustment prospect background is crossed to generate series of results image sequence.
Specific embodiment
To keep the purpose of the present invention, content and advantage clearer, with reference to the accompanying drawing to specific implementation step of the present invention
It is described in further detail.
The present invention devises the multi-source color image moving method based on significant filter, and this method combines digital picture
Treatment theory knowledge compensates for the limited defect of color of individual color image offer, can more effectively realize between color image
Color transfer.Specifically, the present invention comprises the steps of:
S1. characteristics of image is first extracted, then carries out the matching between feature, is obtained and the higher two width shape of shape image similarity
Shape image.SIFT feature extraction process is divided into two processes: first extracting feature offline and stores, then online quick-searching.
Detailed process is as follows: the former is first extracted the SIFT feature of image and stores into database, is later image retrieval procedure
Spare retrieval data are provided.After the completion of off-line procedure, then to target retrieval image carry out SIFT feature extraction, with before from
The database that line is completed carries out characteristic matching, completes on-line retrieval process.Finally, the matched points of statistical picture, matched number
Mesh, and to turn out the image being found in the database more similar to shape image.
S2. super-pixel segmentation is carried out to the obtained color image of step S1 and given shape image.By image from RGB face
For color space transformation to CIE-l α β color space, (l, α, the β) color value and (x, y) coordinate for corresponding to each pixel form one 5 dimension
The similitude of vector V [l, α, β, x, y], two pixels can be measured by their vector distance, and distance is bigger, and similitude is got over
It is small.
Algorithm firstly generates M seed point, and then the detection range seed point is most in the surrounding space of each seed point
Close several pixels, by they be classified as with the seed point one kind, all sort out until all pixels point and finish.Then this K are calculated
The average vector value of all pixels point, retrieves K cluster centre in super-pixel, then again with this K center removal search its
Surrounding and its most similar several pixel, all pixels retrieve K super-pixel after all having sorted out, update cluster centre,
Iteration again, repeatedly until convergence.
S3. brightness is carried out to the obtained color image of step S2 and shape image to remap.During Region Matching, source
The very big situation of the Luminance Distribution difference of image and each pixel of target image often causes large effect to matching result,
Therefore before matching, need first to implement brightness to the channel l of every color image to remap, formula is as follows:
In formulaWithFor the mean value and standard deviation in the channel l in source images s (k)；WithFor the channel target image l
Mean value and variance.For lsNew matching brightness value.
S4. the most matched region pair of color image and shape image is selected.Using significant filter by target image and source
Image segmentation is at foreground and background.The region of high saliency value is known as prospect, and the region of low saliency value is background.In addition, by setting
The segmentation threshold for setting input picture is twice of corresponding notable figure mean value, and picture breakdown Cheng Youqi average color is indicated tight
It gathers, perceives uniform element.By calculating color image region si(k), i=1, the mean value vector of 2 ..., MAnd shape
Image-region tj, the color mean value vector of j=1,2 ..., MBetween Euclidean distance be used as according to being judged, distance is most
Two small regions are best match region pair.The distance between the color mean value in color image and shape image region calculates
Formula is as follows:
By above formula, in that case it can be decided that most matched region is to (tj,si), in tjK candidate regions between siIt is similar source region
Domain is described as follows:
(tj,si)←min({f(tj,si(k)) | 0 < k≤K≤M })
By collecting siIn pixel, we can obtain a compound source images S.
S5. it by the shape image region of the color transfer in each color image region to Corresponding matching, specifically uses such as
Lower formula calculates the transformed value in the channel color image α β:
α in formulatAnd βtIt is target image α, the pixel in the channel β respectively；WithIt is the area target image jKuai respectively
The mean value in the channel the α in domain, β.WithIt is the α in i-th piece of region of color image, the mean value in the channel β respectively；WithPoint
It is not the α in target image jth block region, the standard variance in the channel β；WithIt is the α in i-th piece of region of color image respectively,
The standard variance in the channel β.
The conversion value in the channel color image l is specifically calculated using following formula:
Each Regional Color Transfer result images are synthesized into complete image, then the image of synthesis is converted from the space α β l
Return RGB color.
For a person skilled in the art, after reading above description, various changes and modifications undoubtedly be will be evident.
Therefore those skilled in the art under the inspiration of the present invention, in the ambit for not departing from the claims in the present invention and being protected
Under, replacement or deformation can also be made, is fallen within the scope of protection of the present invention, the range that is claimed of the invention should be with institute
Subject to attached claim.
Detailed description of the invention
The method flow diagram of the method for the present invention.
Claims (4)
1. a kind of multi-source color image color moving method based on significant filter, which is characterized in that the method includes steps
It is rapid:
S1. characteristics of image is first extracted, then carries out the matching between feature, is obtained and the higher two width shape graph of shape image similarity
Picture.SIFT feature extraction process is divided into two processes: first extracting feature offline and stores, then online quick-searching.In detail
Process is as follows: the former is first extracted the SIFT feature of image and stores into database, provides for later image retrieval procedure
Spare retrieval data.After the completion of off-line procedure, SIFT feature extraction then is carried out to target retrieval image, and it is offline before complete
At database carry out characteristic matching, complete on-line retrieval process.Finally, the matched points of statistical picture, matched number are got over
It is more similar to shape image to turn out the image being found in the database more.
S2. super-pixel segmentation is carried out to the obtained color image of step S1 and given shape image.Image is empty from RGB color
Between be transformed into CIE-l α β color space, (l, α, the β) color value and (x, y) coordinate of corresponding each pixel form 5 dimensional vectors
The similitude of V [l, α, β, x, y], two pixels can be measured by their vector distance, and distance is bigger, and similitude is smaller.
Algorithm firstly generates M seed point, and then the detection range seed point is nearest in the surrounding space of each seed point
Several pixels, by they be classified as with the seed point one kind, all sort out until all pixels point and finish.Then this K super pictures are calculated
The average vector value of all pixels point, retrieves K cluster centre in element, then again with this K center removal search around it
With its most similar several pixel, all pixels retrieve K super-pixel after all having sorted out, and update cluster centre, again
Iteration, repeatedly until convergence.
S3. brightness is carried out to the obtained color image of step S2 and shape image to remap.During Region Matching, source images
Very big situation often causes large effect to matching result with the Luminance Distribution difference of each pixel of target image, therefore
Before matching, it needs first to implement brightness to the channel l of every color image to remap, formula is as follows:
In formulaWithFor the mean value and standard deviation in the channel l in source images s (k)；WithFor the equal of the channel target image l
Value and variance.For lsNew matching brightness value.
S4. the most matched region pair of color image and shape image is selected.Using significant filter by target image and source images
It is divided into foreground and background.The region of high saliency value is known as prospect, and the region of low saliency value is background.In addition, defeated by being arranged
The segmentation threshold for entering image is twice of corresponding notable figure mean value, picture breakdown Cheng Youqi average color is indicated compact, sense
Know uniform element.By calculating color image region si(k), i=1, the mean value vector of 2 ..., MWith shape image area
Domain tj, the color mean value vector of j=1,2 ..., MBetween Euclidean distance be used as according to being judged, apart from the smallest by two
A region is best match region pair.The distance between the color mean value in color image and shape image region calculation formula is such as
Under:
By above formula, in that case it can be decided that most matched region is to (tj,si), in tjK candidate regions between siIt is similar source region, retouches
It states as follows:
(tj,si)←min({f(tj,si(k)) | 0 < k≤K≤M })
By collecting siIn pixel, we can obtain a compound source images S.
S5. by the shape image region of the color transfer in each color image region to Corresponding matching, specific use is calculated as follows
The transformed value in the formula calculating channel color image α β:
α in formulatAnd βtIt is target image α, the pixel in the channel β respectively；WithIt is target image jth block region respectively
The mean value in the channel α, β.WithIt is the α in i-th piece of region of color image, the mean value in the channel β respectively；WithIt is respectively
The α in target image jth block region, the standard variance in the channel β；WithIt is the α in i-th piece of region of color image respectively, β is logical
The standard variance in road.
The conversion value in the channel color image l is specifically calculated using following formula:
Each Regional Color Transfer result images are synthesized into complete image, then the image of synthesis is converted back into RGB from the space l α β
Color space.
2. the multi-source color image color moving method according to claim 1 based on significant filter, which is characterized in that
Min ({ f (t is utilized in step 4j,si(k)) | 0 < k≤K≤M }) most matched region is obtained to (tj,si), use Euclidean distance
Judged as foundation, be best match region pair apart from the smallest two regions, is specially calculated using following formula
dij:
3. the multi-source color image color moving method according to claim 2 based on significant filter, which is characterized in that
In step 5, the transformed value in the channel color image l α β is specifically calculated using following formula:
α in formulatAnd βtIt is target image α, the pixel in the channel β respectively；WithIt is target image jth block region respectively
The mean value in the channel α, β.WithIt is the α in i-th piece of region of color image, the mean value in the channel β respectively；WithIt is respectively
The α in target image jth block region, the standard variance in the channel β；WithIt is the α in i-th piece of region of color image respectively, β is logical
The standard variance in road.
4. according to the method described in claim 1, the multi-source color image color migration side based on significant filter newly proposed
Method can generate a series of result images by adjusting the proportionality coefficient of prospect background, select optimal result figure
Picture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811535193.5A CN109710791A (en) | 2018-12-14 | 2018-12-14 | A kind of multi-source color image color moving method based on significant filter |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811535193.5A CN109710791A (en) | 2018-12-14 | 2018-12-14 | A kind of multi-source color image color moving method based on significant filter |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109710791A true CN109710791A (en) | 2019-05-03 |
Family
ID=66256497
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811535193.5A Pending CN109710791A (en) | 2018-12-14 | 2018-12-14 | A kind of multi-source color image color moving method based on significant filter |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109710791A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111080722A (en) * | 2019-12-11 | 2020-04-28 | 中山大学 | Color migration method and system based on significance detection |
| CN111127333A (en) * | 2019-11-12 | 2020-05-08 | 杭州电子科技大学 | Improved color enhancement method for color image facing to bicolor vision |
| CN111476831A (en) * | 2020-03-20 | 2020-07-31 | 清华大学 | PCB image color migration device and method based on cluster analysis |
| CN112907485A (en) * | 2021-03-18 | 2021-06-04 | 国家海洋信息中心 | Remote sensing image batch color matching method based on l alpha beta space color mapping |
| CN113077405A (en) * | 2021-03-27 | 2021-07-06 | 荆门汇易佳信息科技有限公司 | Color transfer and quality evaluation system for two-segment block |
| CN113284058A (en) * | 2021-04-16 | 2021-08-20 | 大连海事大学 | Underwater image enhancement method based on migration theory |
| WO2022052669A1 (en) * | 2020-09-14 | 2022-03-17 | Oppo广东移动通信有限公司 | Background image generation method and apparatus, storage medium, and electronic device |
| CN114627016A (en) * | 2022-03-15 | 2022-06-14 | 厦门微亚智能科技有限公司 | Industrial defect detection preprocessing method based on color migration strategy |
-
2018
- 2018-12-14 CN CN201811535193.5A patent/CN109710791A/en active Pending
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111127333A (en) * | 2019-11-12 | 2020-05-08 | 杭州电子科技大学 | Improved color enhancement method for color image facing to bicolor vision |
| CN111127333B (en) * | 2019-11-12 | 2023-05-05 | 杭州电子科技大学 | Improved color enhancement method for two-color vision-oriented color image |
| CN111080722A (en) * | 2019-12-11 | 2020-04-28 | 中山大学 | Color migration method and system based on significance detection |
| CN111080722B (en) * | 2019-12-11 | 2023-04-21 | 中山大学 | Color migration method and system based on significance detection |
| CN111476831A (en) * | 2020-03-20 | 2020-07-31 | 清华大学 | PCB image color migration device and method based on cluster analysis |
| WO2022052669A1 (en) * | 2020-09-14 | 2022-03-17 | Oppo广东移动通信有限公司 | Background image generation method and apparatus, storage medium, and electronic device |
| CN112907485A (en) * | 2021-03-18 | 2021-06-04 | 国家海洋信息中心 | Remote sensing image batch color matching method based on l alpha beta space color mapping |
| CN112907485B (en) * | 2021-03-18 | 2023-04-25 | 国家海洋信息中心 | Remote sensing image batch toning method based on lalpha beta space color mapping |
| CN113077405A (en) * | 2021-03-27 | 2021-07-06 | 荆门汇易佳信息科技有限公司 | Color transfer and quality evaluation system for two-segment block |
| CN113284058A (en) * | 2021-04-16 | 2021-08-20 | 大连海事大学 | Underwater image enhancement method based on migration theory |
| CN113284058B (en) * | 2021-04-16 | 2024-04-16 | 大连海事大学 | Underwater image enhancement method based on migration theory |
| CN114627016A (en) * | 2022-03-15 | 2022-06-14 | 厦门微亚智能科技有限公司 | Industrial defect detection preprocessing method based on color migration strategy |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109710791A (en) | A kind of multi-source color image color moving method based on significant filter | |
| CN106778604B (en) | Pedestrian re-identification method based on matching convolutional neural network | |
| CN106127791B (en) | A kind of contour of building line drawing method of aviation remote sensing image | |
| CN111709901B (en) | FCM cluster matching and Wallis filtering-based no-weight multi/hyperspectral remote sensing image color homogenizing method | |
| CN109446922B (en) | Real-time robust face detection method | |
| CN106296695A (en) | Adaptive threshold natural target image based on significance segmentation extraction algorithm | |
| CN110706294A (en) | Method for detecting color difference degree of colored textile fabric | |
| CN107123130B (en) | Kernel correlation filtering target tracking method based on superpixel and hybrid hash | |
| CN108537239A (en) | A kind of method of saliency target detection | |
| CN110084782B (en) | Full-reference image quality evaluation method based on image significance detection | |
| CN108600865A (en) | A kind of video abstraction generating method based on super-pixel segmentation | |
| CN110309781A (en) | Damage remote sensing recognition method in house based on the fusion of multi-scale spectrum texture self-adaption | |
| CN108230340A (en) | A kind of SLIC super-pixel extraction Weighting and super-pixel extracting method based on MMTD | |
| CN108846404A (en) | A kind of image significance detection method and device based on the sequence of related constraint figure | |
| CN112801896B (en) | Backlight image enhancement method based on foreground extraction | |
| CN104657980A (en) | Improved multi-channel image partitioning algorithm based on Meanshift | |
| CN113963032A (en) | Twin network structure target tracking method fusing target re-identification | |
| CN103020933A (en) | Multi-source image fusion method based on bionic visual mechanism | |
| CN107169962A (en) | The gray level image fast partition method of Kernel fuzzy clustering is constrained based on space density | |
| CN111179193A (en) | Dermatoscope image enhancement and classification method based on DCNNs and GANs | |
| CN111199245A (en) | Rape pest identification method | |
| CN107481183B (en) | Low-light-level image colorization method based on multi-dimensional data association rule | |
| CN107610137A (en) | A kind of high-resolution remote sensing image optimal cut part method | |
| CN105825234A (en) | Superpixel and background model fused foreground detection method | |
| Zhao et al. | Color channel fusion network for low-light image enhancement |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190503 |