CN102254326A - Image segmentation method by using nucleus transmission - Google Patents
Image segmentation method by using nucleus transmission Download PDFInfo
- Publication number
- CN102254326A CN102254326A CN2011102086183A CN201110208618A CN102254326A CN 102254326 A CN102254326 A CN 102254326A CN 2011102086183 A CN2011102086183 A CN 2011102086183A CN 201110208618 A CN201110208618 A CN 201110208618A CN 102254326 A CN102254326 A CN 102254326A
- Authority
- CN
- China
- Prior art keywords
- seed
- super pixel
- super
- image
- num
- 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
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses an image segmentation method by using nucleus transmission to solve problems of large storage scale and data inconsistency in a present method. The method comprises the following steps: inputting an image, extracting color characteristics of the image, obtaining a super-pixel set of the input image by using a mean value shift method, and calculating a super-pixel color characteristic set; searching a seed point set in the super-pixel color characteristic set by using a k-means clustering method; updating a label of the seed point set by using an adaptive spectral clustering method and forming a constraint set; sending constraint information in the constraint set to whole super-pixel color feature set space by using a nucleus transmission method and obtaining a nucleus matrix; clustering the nucleus matrix by using a k- means method to obtain label vector of the super-pixel, and outputting a segmentation result. The method in the invention has the characteristics of low storage scale, maintenance of data consistency, high calculating efficiency and high segmentation precision, and can be used for object detection and tracking, medical image analysis, network image retrieval and conference video monitoring.
Description
Technical field
The invention belongs to image processing field, relate to a kind of image partition method, specifically a kind of method of utilizing the image segmentation of nuclear transmission can be used for target detection and tracking, medical image analysis, network image retrieval and the monitoring of meeting screen.
Background technology
Digital image processing techniques are fields interdisciplinary.Continuous development along with computer science and technology, Flame Image Process and analysis have formed independently scientific system gradually, image segmentation is a kind of important images treatment technology, can be applicable to medical image and detects focal zone, target recognition and tracking, network image retrieval, video monitoring etc.Image segmentation is the committed step of Flame Image Process, we can say, the quality of image segmentation result directly influences the understanding to image.
The method of image segmentation and kind have a lot, common cutting techniques: Threshold Segmentation technology, differentiating operator rim detection, region growth technique, cluster segmentation technology.The cluster segmentation technology that remerges based on pre-segmentation is a research focus of cutting apart the field in recent years in the world.
The image segmentation that remerges based on pre-segmentation, at first utilize the image partition method of over-segmentation, image is divided into a lot of zones, and these are called super pixel by the zonule that over-segmentation obtains, and adopt sorting technique that these super pixel classification are merged again and obtain best image segmentation result.This method is converted into image segmentation problem regional consolidation problem in essence.Can be reduced in the data volume that to handle when merging greatly because this dividing method is compared with traditional cluster segmentation technology, therefore cause extensive concern in recent years.Tao and Jin proposed color image segmentation method elegant based on average and that standard is cut in 2007 according to this thought, and this method is utilized average drifting over-segmentation image, obtains super pixel, and code requirement suits and surpasses pixel and obtains segmentation result.Ersahin and Cumming proposed a kind of new polarization segmentation method for synthetic aperture radar images based on the spectrogram division in 2010, this method utilizes profile to become the zonule to obtain super pixel image segmentation with spatial information, adopts spectrogram to divide and merges super pixel acquisition segmentation result.Huang and Sang are proposing to utilize the cohesion clustering method to carry out image segmentation in the L*a*b color space at the beginning of 2011, at first utilize Mori to obtain super pixel, adopt the super pixel of cohesion strategy merging to obtain segmentation result then in the method for the super pixel of acquisition of proposition in 2005.Although these methods have improved counting yield, still there are some problems: 1) can not keep the consistance between super pixel count strong point; 2) adopt Gauss's similar function to calculate similarity between the super pixel, must scale parameter be set manually artificially, make the segmentation result instability; 3) only consider local characteristics, lost the global optimization characteristic and reduced segmentation precision.
Summary of the invention
The objective of the invention is to solve above-mentioned the deficiencies in the prior art, a kind of method of utilizing nuclear transmission to carry out image segmentation is proposed, make full use of existing unmarked view data and make that the result of cutting apart is stable, keep the view data consistance, and do not need artificially to be provided with scale parameter, stablize segmentation result, considered the global property of image, improved efficient and the precision cut apart.
Realize that the object of the invention technical scheme is: utilize the image partition method of over-segmentation earlier, image is divided into a lot of zones, these are called super pixel by the zonule that over-segmentation obtains, and adopt the nuclear transmission method that these super pixels are merged again and obtain image segmentation result.Its concrete steps comprise as follows:
1) input piece image, the color characteristic F={f of this image of acquisition in the Luv color space
L, f
u, f
v, the feature of the pixel of each line display of F wherein, f
L, f
u, f
vThe feature of representing luminance component L, chromaticity coordinate component u and the chromaticity coordinate component v of Luv color space respectively;
2) adopting the average drifting method is super set of pixels with this input picture pre-segmentation
The mean value of all pixel features that each super pixel comprised respectively as each super color of pixel feature, is obtained super pixel color feature set
Wherein super pixel s
iBe i zone of average drifting method pre-segmentation input picture gained, i=1,2 ..., n, n represent the size of super set of pixels S, sf
iBe i super pixel s
iColor characteristic;
3) utilize overall k-means cluster, poly-super pixel color feature set SF for the num_seed class, obtain the cluster centre collection
Use Euclidean distance in super pixel color feature set SF, to seek the nearest super pixel color feature corresponding, obtain the seed points set with each cluster centre according to nearest neighbouring rule
Wherein num_seed is the number of the artificial seed points of determining, c
iRepresent i cluster centre, se
iBe the super pixel color feature that cluster centre is nearest corresponding to i, i=1,2 ..., num_seed;
4) adopt the Adaptive spectra cluster the seed points S set
SeedPoly-is the N class, obtains the seed points tally set
If per two seed points have identical label then belong to must-link constrain set M, if having different labels, per two seed points belong to cannot-link constrain set C, wherein N is the artificial image classification number of setting, sl
iThe label of representing i seed points, sl
i∈ 1,2 ..., and N}, i=1,2 ..., hum_seed;
5) method of using nuclear to transmit is delivered to the paired constraint information among must-link constrain set M and the cannot-link constrain set C among the whole super pixel color feature set SF, obtains size and is the nuclear matrix K of n*n;
6) adopt the k-means clustering method, poly-for the N class nuclear matrix K by row, obtain cluster label vector
L wherein
iRepresent i super pixel s
iBe split into l
iClass, l
i∈ 1,2 ..., N};
7) will the super pixel s of same label be arranged in the image
iDistribution is exported the image after cutting apart with a kind of color.
The present invention has following characteristics compared with prior art:
1. the present invention adopts the average drifting pre-segmentation to obtain super set of pixels, split image on the basis of super pixel, effectively kept each target the edge, greatly reduce view data to be processed, improved counting yield.
2. the present invention adopts overall k-means clustering method, seeks seed points in whole data space, makes seed points have more representativeness.
3. the present invention adopts the Adaptive spectra clustering method to obtain must-link constrain set and cannot-link constrain set, does not need artificially to be provided with scale parameter, obtains stable segmentation result.
4. the present invention adopts the method for nuclear transmission that constraint information is delivered to whole data space, has effectively kept the consistance of data, has improved segmentation precision.
The simulation experiment result shows that the present invention utilizes the nuclear transmission method effectively to be used for image segmentation, and the understanding of successive image and decipher can better be carried out.
Description of drawings
Fig. 1 is a process flow diagram of the present invention;
Fig. 2 is the original test pattern that uses in the l-G simulation test of the present invention, airplane, horse, flower and church;
Fig. 3 is existing method and the segmentation result of the present invention on four width of cloth test patterns.
Embodiment
With reference to Fig. 1, performing step of the present invention comprises as follows:
The input piece image extracts the color characteristic of this image at the Luv color space, and the color characteristic of all pixels constitutes the matrix F={ f that size is num_pixel*3
L, f
u, f
v, its each row is represented a color of pixel feature, and num_pixel represents the pixel number of this image, f
L, f
u, f
vThe feature of representing luminance component L, chromaticity coordinate component u and the chromaticity coordinate component v of Luv color space respectively.
Step 2, the pre-segmentation input picture obtains super set of pixels
(2.1) adopt average drifting method pre-segmentation input picture, obtain the label of each pixel, label range is sought the pixel with same label from 1 to n, and, formulate a label s for each overdivided region there being the pixel of same label to merge into a zone
i, i=1,2 ..., n, overdivided region are called super pixel, obtain the super set of pixels of input picture
N represents the size of super set of pixels S.This step also can adopt dividing ridge method to obtain the super set of pixels of input picture, the average drifting method is referring to document: D.Comaniciu, P.Meer, " Mean shift:a robust approach toward feature space analysis; " IEEE Transactions on Pattern Analysis and Machine Intelligence, vol.24, no.5, pp.603-619,2002;
(2.2) get the mean value of the color characteristic of the pixel that each super pixel comprises, with this mean value as the super pixel s of correspondence
iColor characteristic sf
i, constitute size and be the super pixel color feature set of n*3
Step 3 is sought seed points
(3.1) use overall k-means clustering method, poly-super pixel color feature set SF for the num_seed class, obtain the cluster centre collection
Wherein num_seed people is the number of definite seed points, c
iRepresent i cluster centre, i=1,2 ..., num_seed.Overall situation k-means clustering method is referring to document: A.Likas, N.Vlassis, and and J.J.Verbeek, " The global clustering algorithm, " Pattern Recognition, vol.36, no.2, pp.451-4-61,2003;
(3.2) each row among the super pixel color feature set SF is regarded as a point, simultaneously the cluster centre collection
Each row is regarded a point as, according to the Euclidean distance calculating formula
Calculate the Euclidean distance of each super pixel color feature, sf to all cluster centres
iAnd c
jThree-dimensional feature is arranged respectively, use num_seed the Euclidean distance that calculates as distance matrix d=[d
Ij]
N*num_seedI capable, i=1,2 ..., n, j=1,2 ..., num_seed;
(3.3) in each row of distance matrix d, seek minimum value, and note the corresponding rower of this value and sign, in super pixel color feature set SF, take out the row of this label correspondence, constitute the seed points set
I=1,2 ..., num_seed.
Step 4, structure constrain set
(4.1) adopt the Adaptive spectra cluster with the seed points S set
SeedPoly-for the N class, obtain the seed points tally set
Wherein N is the artificial image classification number of setting, sl
iThe label of representing i seed points, sl
i∈ 1,2 ..., and N}, i=1,2 ..., num_seed.The Adaptive spectra clustering method is referring to document: L.Zelnik-Manor and P.Perona, " Self-tuning spectral clustering ", in:Proceedings of Advances in Neural Information Processing, 2004;
(4.2) check current seed points se
iWith next seed points se
I+1Seed points label sl
iWith sl
I+1Whether identical, if, then se
iWith se
I+1Location records corresponding in super pixel color feature set SF advances must-link constrain set M, its line number adds 1 simultaneously, otherwise then record advances cannot-link constrain set C, its line number adds 1 simultaneously, and the line number sum of must-link constrain set M and cannot-link constrain set C is num_seed-1, and columns is respectively 2, i=1,2 ..., num_seed-1.
Step 5 is transmitted constraint information and is obtained nuclear matrix
Must-link constrain set M that will be obtained by step 4 and the paired constraint information of cannot-link constrain set C and super pixel color feature set SF are mapped to a nuclear space according to the mapping relations of nuclear transmission method, obtain size and are the nuclear matrix K of n*n.The nuclear transmission method is referring to document: E.Hu, S.Chen, D.Zhang, and X.Yin, " Semisupervised kernel matrix learning by kernel propagation, " IEEE Transactionson Neural Networks, vol.21, no.11, pp.1831-1841,2010.
Step 6 obtains super set of pixels and gets the label vector
Each row of the nuclear matrix K of step 5 gained is regarded as a data point, and then total n data point with the input of this n data point as the k-means clustering method, obtains the label vector of this n data point
L wherein
iRepresent that i data point is l by gathering
iClass, and i super pixel s
iBe split to l
iClass, l
i∈ 1,2 ... and N}, i=1,2 ..., n.The k-means clustering procedure is referring to document: G.P.Babu and M.N.Murty; " Simulated annealing for selecting initialseeds in the k-means algorithm, " Ind.J.Pure Appl.Math, vol.25; pp.85-94,1994.
Step 7, the image after output is cut apart
With the super pixel segmentation of same label in the image is same class, and every class is distributed different colors, obtains split image, the image after output is cut apart.
Effect of the present invention can further specify by following experiment:
1. simulated conditions:
At CPU is pentium (R) 4,1.86GHZ, internal memory 2G, WINDOWS XP system, has carried out emulation on the Matlab7.10 platform.
2. emulation content:
(2.1) the present invention verifies its effect on Berkeley segmentation database.The method of doing to contrast with the present invention has traditional clustering method: k-means clustering method (KM) and based on
The spectral clustering method (NSC) of approaching, and the method that remerges of the development in recent years pre-segmentation of getting up: the color image segmentation method of cutting based on average drifting and standard (MSNC) and based on compressed texture merging method (CTM).Used appraisal procedure is: edge displacement error rate (BDE), and minimum value is 0, and is the smaller the better; Similar probability index (PRI), scope is the bigger the better between 0 and 1; Information change rate (VoI), minimum value are 0, are worth the smaller the better; The overall situation consistent error rate (GCE), scope is between 0 and 1, and is the smaller the better.Table 1 has shown the average assessment result of these five kinds of methods on 30 width of cloth Berkeley images.Black matrix is represented best.As can be seen from Table 1, the present invention is better than other four kinds of methods greatly on BDE, also good than other four kinds of methods on VoI, on PRI than best values little 0.0093, on GCE than best values height 0.0491, take all factors into consideration the present invention and be better than other four kinds of methods.
Table 1: the present invention and the existing method average result contrast on 30 width of cloth Berkeley images
| Method | BDE | PRI | VoI | GCE |
| KM | 6.6421 | 0.8002 | 2.1662 | 0.2630 |
| NSC | 7.5890 | 0.7676 | 2.4270 | 0.3123 |
| MSNC | 6.3071 | 0.7647 | 2.2350 | 0.3387 |
| CTM | 6.6160 | 0.8073 | 2.0596 | 0.1803 |
| The present invention | 5.4283 | 0.7980 | 2.0478 | 0.2294 |
(2.2) selected four width of cloth images among Fig. 2 as showing at Berkeley segmentation database, and the present invention and KM and MSNC have been contrasted.Four width of cloth test patterns are respectively the church images shown in the flower image shown in the horse image shown in the airplane image shown in Fig. 2 (a), Fig. 2 (b), Fig. 2 (c) and Fig. 2 (d).To airplane image num_seed=100, N=2, to horse image num_seed=100, N=2, to flower image num_seed=100, N=3, to church image num_seed=200, N=10.
Airplane imagery exploitation the present invention shown in Fig. 2 (a) and KM and MSNC method are carried out image segmentation, and the result is respectively as Fig. 3 (a), Fig. 3 (b) and Fig. 3 (c).The present invention has effectively kept the edge of aircraft and the consistance of fuselage data as can be seen from Fig. 3 (a), and experiment shows that segmentation result is stable; Fig. 3 (b) and Fig. 3 (c) demonstrate the marginal information that KM and MSNC have lost aircraft, and the homogeneous region consistance is poor, and experiment shows the segmentation result instability.
Horse imagery exploitation the present invention shown in Fig. 2 (b) and KM and MSNC method are carried out image segmentation, and the result is respectively as Fig. 3 (d), Fig. 3 (e) and Fig. 3 (f).The separatrix on two dry goods and meadow is clear as can be seen from Fig. 3 (d), and homogeneous region keeps good consistance, does not have noise spot; Fig. 3 (e) demonstrates the horse that KM cuts apart good edge, but there is noise spot in the meadow homogeneous region; Fig. 3 (f) demonstrates horse and the meadow that MSNC cuts apart and mixes repeatedly, and a meadow part is cut apart and the same class of horse.
Flower imagery exploitation the present invention shown in Fig. 2 (c) and KM and MSNC method are carried out image segmentation, and the result is respectively as Fig. 3 (g), Fig. 3 (h) and Fig. 3 (i).The present invention is separated with pistil petal and greenery clearly as can be seen from 3 (g), and it is good that marginal information keeps, and kept the homogeneous region consistance; Pistil information that Fig. 3 (h) and 3 (i) have demonstrated missing image that KM and MSNC cut apart is divided into a class with pistil and greenery, and the consistance of greenery homogeneous region is very poor simultaneously.
Church imagery exploitation the present invention shown in Fig. 2 (d) and KM and MSNC method are carried out image segmentation, and the result is respectively as Fig. 3 (j), Fig. 3 (k) and Fig. 3 (l).The present invention is obviously separated with church and sky as can be seen from Fig. 3 (j), and demonstrates the doorframe in church clearly, has well kept data message, and homogeneous region does not have obvious noise spot; The obvious noise piece appears in the upper right corner that Fig. 3 (k) demonstrates the sky that KM cuts apart, and has lost the consistance of sky homogeneous region; Fig. 3 (l) demonstrates the doorframe edge fog in the church that MSNC cuts apart, and loses detailed information, and the shade in the lower right corner is split into sky, the wrong branch occur.
Claims (4)
1. one kind is utilized nuclear to transmit the method carry out image segmentation, and its feature comprises the steps:
1) input piece image, the color characteristic F={f of this image of acquisition in the Luv color space
L, f
u, f
v, the feature of the pixel of each line display of F wherein, f
L, f
u, f
vThe feature of representing luminance component L, chromaticity coordinate component u and the chromaticity coordinate component v of Luv color space respectively;
2) adopting the average drifting method is super set of pixels with this input picture pre-segmentation
The mean value of all pixel features that each super pixel comprised respectively as each super color of pixel feature, is obtained super pixel color feature set
Wherein super pixel s
iBe i zone of average drifting method pre-segmentation input picture gained, i=1,2 ..., n, n represent the size of super set of pixels S, sf
iBe i super pixel s
iColor characteristic;
3) utilize overall k-means cluster, poly-super pixel color feature set SF for the num_seed class, obtain the cluster centre collection
Use Euclidean distance in super pixel color feature set SF, to seek the nearest super pixel color feature corresponding, obtain the seed points set with each cluster centre according to nearest neighbouring rule
Wherein num_seed is the number of the artificial seed points of determining, c
iRepresent i cluster centre, se
iBe the super pixel color feature that cluster centre is nearest corresponding to i, i=1,2 ..., num_seed;
4) adopt the Adaptive spectra cluster the seed points S set
SeedPoly-is the N class, obtains the seed points tally set
If per two seed points have identical label then belong to must-link constrain set M, if having different labels, per two seed points belong to cannot-link constrain set C, wherein N is the artificial image classification number of setting, sl
iThe label of representing i seed points, sl
i∈ 1,2 ..., and N}, i=1,2 ..., num_seed;
5) method of using nuclear to transmit is delivered to the paired constraint information among must-link constrain set M and the cannot-link constrain set C among the whole super pixel color feature set SF, obtains size and is the nuclear matrix K of n*n;
6) adopt the k-means clustering method, poly-for the N class nuclear matrix K by row, obtain cluster label vector
L wherein
iRepresent i super pixel s
iBe split into l
iClass, l
i∈ 1,2 ..., N};
7) will the super pixel s of same label be arranged in the image
iDistribution is exported the image after cutting apart with a kind of color.
2. utilization nuclear according to claim 1 transmits the method for carrying out image segmentation, it is characterized in that the described use Euclidean distance of step 3) is according to nearest neighbouring rule, in super pixel color feature set SF, seek the nearest super pixel color feature corresponding, carry out as follows with each cluster centre:
2a) each row among the super pixel color feature set SF is regarded as a point, simultaneously the cluster centre collection
Each the row regard a point as, according to the Euclidean distance calculating formula
Calculate the Euclidean distance of each super pixel color feature, sf in the calculating formula to all cluster centres
iAnd c
jThree-dimensional feature is arranged respectively, use num_seed the Euclidean distance that calculates as distance matrix d=[d
Ij]
N*num_seedI capable, i=1,2 ..., n, j=1,2 ..., num_seed;
2b) in each row of distance matrix d, seek minimum value, and note the corresponding rower of this value and sign, in super pixel color feature set SF, take out the row of this label correspondence, constitute the seed points set
I=1,2 ..., num_seed.
3. utilization nuclear according to claim 1 transmits the method for carrying out image segmentation, if it is characterized in that described per two seed points of step 4) has identical label then belongs to must-link constrain set M, if per two seed points have different labels then belong to cannot-link constrain set C, are to check current seed points se
iWith next seed points se
I+1Seed points label sl
iWith sl
I+1Whether identical, if, then se
iWith se
I+1Must-link constrain set M is advanced as a line item in position corresponding in super pixel color feature set SF, otherwise then record advances cannot-link constrain set C, the line number sum of must-link constrain set M and cannot-link constrain set C is num_seed-1, columns is respectively 2, i=1,2 ..., num_seed-1.
4. utilization nuclear according to claim 1 transmits the method for carrying out image segmentation, it is characterized in that the method that the described use nuclear of step 5) transmits is delivered to the paired constraint information among must-link constrain set M and the cannot-link constrain set C among the whole super pixel color feature set SF, be that constraint information and super pixel color feature set are mapped to a nuclear space according to the mapping relations of examining transmission method, obtain nuclear matrix K.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011102086183A CN102254326A (en) | 2011-07-22 | 2011-07-22 | Image segmentation method by using nucleus transmission |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011102086183A CN102254326A (en) | 2011-07-22 | 2011-07-22 | Image segmentation method by using nucleus transmission |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102254326A true CN102254326A (en) | 2011-11-23 |
Family
ID=44981567
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011102086183A Pending CN102254326A (en) | 2011-07-22 | 2011-07-22 | Image segmentation method by using nucleus transmission |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102254326A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663369A (en) * | 2012-04-20 | 2012-09-12 | 西安电子科技大学 | Human motion tracking method on basis of SURF (Speed Up Robust Feature) high efficiency matching kernel |
| CN102663757A (en) * | 2012-04-20 | 2012-09-12 | 西安电子科技大学 | Semi-automatic image cutting method based on nuclear transfer |
| CN102855624A (en) * | 2012-07-23 | 2013-01-02 | 武汉大学 | Image segmentation method based on generalized data field and Normalized cut (Ncut) algorithm |
| CN103218823A (en) * | 2013-05-08 | 2013-07-24 | 西安电子科技大学 | Remote sensing image change detection method based on nuclear transmission |
| CN103324931A (en) * | 2012-03-22 | 2013-09-25 | 北京百度网讯科技有限公司 | Method and device for image color binarization |
| CN103353987A (en) * | 2013-06-14 | 2013-10-16 | 山东大学 | Superpixel segmentation method based on fuzzy theory |
| CN103366382A (en) * | 2013-07-04 | 2013-10-23 | 电子科技大学 | Active contour tracing method based on superpixel |
| CN103679719A (en) * | 2013-12-06 | 2014-03-26 | 河海大学 | Image segmentation method |
| CN105303546A (en) * | 2014-06-20 | 2016-02-03 | 江南大学 | Affinity propagation clustering image segmentation method based on fuzzy connectedness |
| CN106462986A (en) * | 2014-06-12 | 2017-02-22 | 汤姆逊许可公司 | Methods and systems for color processing of digital images |
| CN106709517A (en) * | 2016-12-19 | 2017-05-24 | 航天恒星科技有限公司 | Mangrove recognition method and system |
| CN107657276A (en) * | 2017-09-15 | 2018-02-02 | 赤峰学院 | A kind of Weakly supervised semantic segmentation method based on searching semantic category cluster |
| CN108320283A (en) * | 2018-02-01 | 2018-07-24 | 研靖信息科技（上海）有限公司 | A kind of method and apparatus of the optimal color matching threshold range of determining image |
| CN108921853A (en) * | 2018-06-22 | 2018-11-30 | 西安电子科技大学 | Image partition method based on super-pixel and clustering of immunity sparse spectrums |
| CN110097071A (en) * | 2019-03-18 | 2019-08-06 | 深圳蓝韵医学影像有限公司 | The recognition methods in the breast lesion region based on spectral clustering in conjunction with K-means and device |
| CN110827309A (en) * | 2019-11-12 | 2020-02-21 | 太原理工大学 | Polaroid appearance defect segmentation method based on super-pixels |
| CN111415360A (en) * | 2020-03-30 | 2020-07-14 | 上海烟草集团有限责任公司 | Tobacco leaf image cutting method, device, equipment and medium |
| CN113489896A (en) * | 2021-06-25 | 2021-10-08 | 中国科学院光电技术研究所 | Video image stabilization method capable of robustly predicting global motion estimation |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101226633A (en) * | 2008-01-30 | 2008-07-23 | 哈尔滨工程大学 | Method for segmentation of corps canopy image based on average dispersion |
| CN101251893A (en) * | 2008-03-28 | 2008-08-27 | 西安电子科技大学 | Self-adapting multi-dimension veins image segmenting method based on wavelet and average value wander |
| CN101751666A (en) * | 2009-10-16 | 2010-06-23 | 西安电子科技大学 | Semi-supervised multi-spectral remote sensing image segmentation method based on spectral clustering |
| CN102024262A (en) * | 2011-01-06 | 2011-04-20 | 西安电子科技大学 | Method for performing image segmentation by using manifold spectral clustering |
-
2011
- 2011-07-22 CN CN2011102086183A patent/CN102254326A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101226633A (en) * | 2008-01-30 | 2008-07-23 | 哈尔滨工程大学 | Method for segmentation of corps canopy image based on average dispersion |
| CN101251893A (en) * | 2008-03-28 | 2008-08-27 | 西安电子科技大学 | Self-adapting multi-dimension veins image segmenting method based on wavelet and average value wander |
| CN101751666A (en) * | 2009-10-16 | 2010-06-23 | 西安电子科技大学 | Semi-supervised multi-spectral remote sensing image segmentation method based on spectral clustering |
| CN102024262A (en) * | 2011-01-06 | 2011-04-20 | 西安电子科技大学 | Method for performing image segmentation by using manifold spectral clustering |
Non-Patent Citations (1)
| Title |
|---|
| 王朝英，李光，侯志强，林南章: "融合mean shift 和区域显著性的彩色图像分割算法", 《计算机工程与应用》 * |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103324931A (en) * | 2012-03-22 | 2013-09-25 | 北京百度网讯科技有限公司 | Method and device for image color binarization |
| CN102663757A (en) * | 2012-04-20 | 2012-09-12 | 西安电子科技大学 | Semi-automatic image cutting method based on nuclear transfer |
| CN102663369A (en) * | 2012-04-20 | 2012-09-12 | 西安电子科技大学 | Human motion tracking method on basis of SURF (Speed Up Robust Feature) high efficiency matching kernel |
| CN102855624A (en) * | 2012-07-23 | 2013-01-02 | 武汉大学 | Image segmentation method based on generalized data field and Normalized cut (Ncut) algorithm |
| CN103218823A (en) * | 2013-05-08 | 2013-07-24 | 西安电子科技大学 | Remote sensing image change detection method based on nuclear transmission |
| CN103218823B (en) * | 2013-05-08 | 2016-04-13 | 西安电子科技大学 | Based on the method for detecting change of remote sensing image that core is propagated |
| CN103353987B (en) * | 2013-06-14 | 2015-10-28 | 山东大学 | A kind of superpixel segmentation method based on fuzzy theory |
| CN103353987A (en) * | 2013-06-14 | 2013-10-16 | 山东大学 | Superpixel segmentation method based on fuzzy theory |
| CN103366382A (en) * | 2013-07-04 | 2013-10-23 | 电子科技大学 | Active contour tracing method based on superpixel |
| CN103679719A (en) * | 2013-12-06 | 2014-03-26 | 河海大学 | Image segmentation method |
| CN106462986A (en) * | 2014-06-12 | 2017-02-22 | 汤姆逊许可公司 | Methods and systems for color processing of digital images |
| CN105303546A (en) * | 2014-06-20 | 2016-02-03 | 江南大学 | Affinity propagation clustering image segmentation method based on fuzzy connectedness |
| CN105303546B (en) * | 2014-06-20 | 2018-08-17 | 江南大学 | Neighbour's propagation clustering image partition method based on fuzzy connectedness |
| CN106709517A (en) * | 2016-12-19 | 2017-05-24 | 航天恒星科技有限公司 | Mangrove recognition method and system |
| CN106709517B (en) * | 2016-12-19 | 2020-09-04 | 航天恒星科技有限公司 | Mangrove forest identification method and system |
| CN107657276A (en) * | 2017-09-15 | 2018-02-02 | 赤峰学院 | A kind of Weakly supervised semantic segmentation method based on searching semantic category cluster |
| CN108320283A (en) * | 2018-02-01 | 2018-07-24 | 研靖信息科技（上海）有限公司 | A kind of method and apparatus of the optimal color matching threshold range of determining image |
| CN108921853A (en) * | 2018-06-22 | 2018-11-30 | 西安电子科技大学 | Image partition method based on super-pixel and clustering of immunity sparse spectrums |
| CN108921853B (en) * | 2018-06-22 | 2022-03-04 | 西安电子科技大学 | Image segmentation method based on super-pixel and immune sparse spectral clustering |
| CN110097071A (en) * | 2019-03-18 | 2019-08-06 | 深圳蓝韵医学影像有限公司 | The recognition methods in the breast lesion region based on spectral clustering in conjunction with K-means and device |
| CN110827309A (en) * | 2019-11-12 | 2020-02-21 | 太原理工大学 | Polaroid appearance defect segmentation method based on super-pixels |
| CN110827309B (en) * | 2019-11-12 | 2023-06-23 | 太原理工大学 | Super-pixel-based polaroid appearance defect segmentation method |
| CN111415360A (en) * | 2020-03-30 | 2020-07-14 | 上海烟草集团有限责任公司 | Tobacco leaf image cutting method, device, equipment and medium |
| CN111415360B (en) * | 2020-03-30 | 2022-12-20 | 上海烟草集团有限责任公司 | Tobacco leaf image cutting method, device, equipment and medium |
| CN113489896A (en) * | 2021-06-25 | 2021-10-08 | 中国科学院光电技术研究所 | Video image stabilization method capable of robustly predicting global motion estimation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102254326A (en) | Image segmentation method by using nucleus transmission | |
| CN110399909B (en) | Hyperspectral image classification method based on label constraint elastic network graph model | |
| CN108573276B (en) | Change detection method based on high-resolution remote sensing image | |
| CN102663757A (en) | Semi-automatic image cutting method based on nuclear transfer | |
| CN108898145A (en) | A kind of image well-marked target detection method of combination deep learning | |
| CN107909015A (en) | Hyperspectral image classification method based on convolutional neural networks and empty spectrum information fusion | |
| CN105608692B (en) | Polarization SAR image segmentation method based on deconvolution network and sparse classification | |
| CN107145889A (en) | Target identification method based on double CNN networks with RoI ponds | |
| CN104268583A (en) | Pedestrian re-recognition method and system based on color area features | |
| CN102651128A (en) | Image set partitioning method based on sampling | |
| CN104850822B (en) | Leaf identification method under simple background based on multi-feature fusion | |
| CN104077605A (en) | Pedestrian search and recognition method based on color topological structure | |
| CN109448015A (en) | Image based on notable figure fusion cooperates with dividing method | |
| CN108629783A (en) | Image partition method, system and medium based on the search of characteristics of image density peaks | |
| CN101968852A (en) | Entropy sequencing-based semi-supervision spectral clustering method for determining clustering number | |
| CN103473545A (en) | Text-image similarity-degree measurement method based on multiple features | |
| CN103530882A (en) | Improved image segmentation method based on picture and color texture features | |
| CN103309982A (en) | Remote sensing image retrieval method based on vision saliency point characteristics | |
| CN104408733A (en) | Object random walk-based visual saliency detection method and system for remote sensing image | |
| CN104216974A (en) | Unmanned aerial vehicle aerial image matching method based on vocabulary tree blocking and clustering | |
| CN103678552A (en) | Remote-sensing image retrieving method and system based on salient regional features | |
| CN105243387A (en) | Open-pit mine typical ground object classification method based on UAV image | |
| CN109447111A (en) | A kind of remote sensing supervised classification method based on subclass training sample | |
| Ma et al. | Towards improved accuracy of UAV-based wheat ears counting: A transfer learning method of the ground-based fully convolutional network | |
| CN107657276B (en) | Weak supervision semantic segmentation method based on searching semantic class clusters |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20111123 |