CN102982539A - Characteristic self-adaption image common segmentation method based on image complexity - Google Patents
Characteristic self-adaption image common segmentation method based on image complexity Download PDFInfo
- Publication number
- CN102982539A CN102982539A CN201210448129XA CN201210448129A CN102982539A CN 102982539 A CN102982539 A CN 102982539A CN 201210448129X A CN201210448129X A CN 201210448129XA CN 201210448129 A CN201210448129 A CN 201210448129A CN 102982539 A CN102982539 A CN 102982539A
- Authority
- CN
- China
- Prior art keywords
- image
- divided
- complexity
- cut
- segmentation
- 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
Images
Abstract
The invention provides a characteristic self-adaption image common segmentation method based on image complexity. The method includes in consideration of accurate detection results of an object detection method in a simple background image and dispersing detection results in a complex image and in consideration of initial segmentation accuracy, firstly starting from segmentation results of the simple background image, then self-adaption adjusting a common segmentation parameter by learning corresponding optimal similarity level measurement criteria of different image groups, and using the common segmentation parameter which is obtained by self-adaption adjusting to carry out image common segmentation handling. The method is high in detection rate, accurate in result and strong in self-adaption ability.
Description
Technical field
The present invention relates to image processing techniques, particularly adapting to image is divided into the technology of cutting.
Background technology
Be accompanied by the development of network, exist a large amount of digital pictures in the network.Utilize the image of these magnanimity to realize the discovery of special object and cut apart more and more becoming people's problems of concern.
Be divided into and cut from several different background, comprise the technology that in the image of identical special object special object is split.In order to be partitioned into the special object in the present image, the existing segmentation method that is divided into is at first introduced the image that several comprise identical special object, such as the image that search engine obtains, then realize cutting apart of special object by extracting and be partitioned into the shared object that comprises in this group image.
Be divided into the advantage of segmentation method for only needing the user to introduce assistant images, user's degree of participation is low, and workload is less.At present, existing multiple digital picture is divided into segmentation method and is suggested, be divided into segmentation method such as the image based on markov random file, be divided into segmentation method based on the image of differentiating cluster, image based on the thermal diffusion theory is divided into segmentation method, image based on the active profile is divided into segmentation method, is divided into segmentation method and is divided into segmentation method etc. based on the image of shortest path based on the image of random walk.In these methods, image is divided into the problem of cutting and usually is characterized as optimization problem.This optimization problem is considered two aspects, the one, the Object Segmentation of single image, the i.e. flatness of the difference of prospect and background and regional area pixel tag; The 2nd, the similarity degree of prospect between different images, namely requiring the result of cutting apart is shared object.The Object Segmentation of single image realizes by traditional single image dividing method usually, and shared object is cut apart then and joined in the single image parted pattern as add-ins, by energy term corresponding to optimization, shared object is split.
A key problem that is divided in the segmentation method is how to weigh the similarity degree of foreground area.If the selection of similarity degree is unreasonable, just shared object can't split from object.Existing method usually adopts fixing similarity degree criterion, namely is being divided into when cutting, even for the shared object of different images group, system uses identical a series of characteristic parameters that cut that are divided into are set.As with people's face as the image sets of shared object with aircraft is divided into all not variations of parameter such as the color of cutting, shape, texture as the image sets of shared object is employed.In fact, the same characteristic features that shared object is corresponding in the different images group varies, and is divided into parameter such as the untimely adjustment of cutting and impact is divided into the accuracy of cutting the result.But be divided into segmentation method as automatic image, people often can't participate in the selection of characteristic parameter, are difficult to the characteristic parameter that participates in cutting apart is manually adjusted.
Summary of the invention
Technical matters to be solved by this invention is, provides a kind of and can the self-adaptation adjustment be divided into the image that cuts characteristic parameter and be divided into segmentation method.
The present invention solves the problems of the technologies described above the technical scheme that adopts to be, a kind of feature adapting to image based on the image complexity analysis is divided into segmentation method, may further comprise the steps:
1) complexity of each image in the image sets of calculating input is chosen m the simple image of background in the image sets of input; M is the positive integer greater than 1;
2) m the simple image initial of background is partitioned into separately foreground area, obtains initial segmentation result;
3) utilizing initial segmentation result and initially being divided into to cut parameter obtains by the expectation maximization learning algorithm so that have the optimum of optimum similarity between the segmentation result of m the simple image of background and be divided into and cut parameter vector;
4) utilize optimum being divided into to cut parameter vector and all images in the input picture group are divided into to cut realize cutting apart shared object in all images.
The present invention considers that the testing result of method for checking object in the simple background image is more accurate, and the testing result in the complicated image is disperseed, consider the initial segmentation accuracy, at first from the segmentation result of simple background image, come the self-adaptation adjustment to be divided into by optimum similarity degree criterion corresponding to study different images group again and cut parameter.
The invention has the beneficial effects as follows, be divided into based on the self-adaptation adjustment of image complexity analysis and cut parameter that the segmentation method verification and measurement ratio is high so that image of the present invention is divided into, the result is accurate, adaptive ability is strong.
Description of drawings
Fig. 1 cuts the result according to being divided into of different images group that the present embodiment method obtains.
Embodiment
Present embodiment is realized at Matlab R2010a experiment porch, mainly comprises three steps, is respectively the generation with initial segmentation of obtaining of simple image, and the study of characteristic model and be divided into and cut realization is specific as follows:
Step 1, simple image obtain generation with initial segmentation.Specifically comprise following substep:
The 1st step: the complicacy of analysis image and ordering.By the computational complexity mark
With the complexity mark
Realize.
1. adopt multiple dimensioned image over-segmentation method based on the edge with image I
i, i=1 ..., N
iToo be slit into regional area, ask for image complexity mark
Over-segmentation refers to that with image segmentation be a plurality of regional areas, and the semanteme of consideration of regional not:
Based on image I
iCorresponding complexity mark
Size each image is sorted, obtain ranking results, distribute image I
iSequence number after the corresponding ordering
2. consider that the testing result of method for checking object is more accurate in the simple image, and the detection in the complicated image added dispersion.Employing comprises the highest front N of prospect probable value based on what the object detection method of moving window was obtained
wIndividual window.Each window is expressed as a two values matrix M
k, k=1 ..., N
wWherein numerical value corresponding to the pixel in the window is 1, and outer numerical value corresponding to pixel of window is 0, with all matrix M
kStack obtains M:
Ask for corresponding complexity mark mark
That is:
Wherein
T
wThe expression threshold value, M (j, k) is in image I
iMiddle employing comprises the highest front N of prospect probable value based on what the object detection method of moving window was obtained
wIndividual window judges that pixel (j, k) is the quantity of prospect.Based on image I
iCorresponding complexity mark
Size each image is sorted, obtain ranking results, distribute image I
iSequence number after the corresponding ordering
3. image I
iCorresponding final ranking results is
Those skilled in the art can easily expect using the existing computing method that other can embody the image complexity to realize the calculating of image complexity.
The 2nd step: the final ranking results taking-up complexity according to each image in the image sets is low that front m image is the simple image of background, by the Object Segmentation method based on remarkable detection m the simple image initial of background is cut apart.The initial segmentation here is not limited to the Object Segmentation method based on remarkable detection, and those skilled in the art can easily expect processing this step with other existing Object Segmentation method.
Step 2, based on initial segmentation result, by expectation maximization EM learning algorithm study optimal characteristics model.Comprise following substep:
2-1, read default being divided into and cut parameter vector; Being divided into and cutting parameter vector θ is all characteristic parameter ω
1..., ω
nThe transposition θ of the row vector that forms=(ω
1..., ω
n)
T, namely being divided into and cutting parameter vector is a column vector that length is n;
2-2, cut parameter vector θ and current segmentation result is asked for current similarity matrix S (X, θ) according to current being divided into; Initially current is divided into and cuts parameter vector θ is that parameter vector is cut in default being divided into, and initial current segmentation result X is initial segmentation result;
S (X, θ)=X θ=(X
1, X
2..., X
n) θ, wherein, n is the total number of feature, the segmentation result of each width of cloth image is asked for n feature;
P, q=1 ..., m, l=1 ..., n, m represent to cut apart quantity;
D () representation feature distance function,
Be image I
pL feature in the foreground area that is partitioned into,
Be image I
qL the feature that is partitioned into;
Any two image I
p, I
qSegmentation result between similarity degree be
Total m initial segmentation result cut apart and can be used one group for any two
Represent, total m*m kind combination, m*m is capable altogether for matrix X, the array that a pair of segmentation result of each line display (being obtained by two segmentation results) is corresponding, the length of array is n, namely matrix X is total to the n row; Similarity matrix S (X, θ) also is the matrix of a mm*n;
2-3, ask for current degree of confidence vector Z: the Z=V/max (V) of cutting apart based on similarity matrix S (X, θ); Wherein, V={v
1, v
2..., v
mFor length is the column vector of m,
P=1 ..., m,
Expression among the similarity matrix S with image I
pAll relevant row of segmentation result sue for peace; M segmentation result respectively corresponding degree of confidence of cutting apart is Z[Z (1), Z (2) ..., Z (m)], degree of confidence is between [0,1];
2-4, based on the current degree of confidence vector Z of cutting apart, ask for observation data X
New: X
New(r, t)=X (r, t) S
"" (r); 1≤r≤mm; 1≤t≤n; X (r, t) represents that the position is at the numerical value of (r, t) in the current segmentation result matrix; S " the delegation of " (r)=min (Z (p); Z (q)), S " " (r) represents objective matrix S ", the capable data of r are by two image I of correspondence
p, I
qThe degree of confidence of cutting apart obtain, Z (p) is for cutting apart the capable data of degree of confidence vector Z p, Z (q) is for cutting apart the capable data of degree of confidence vector Z q, the matrix S that obtains " " (r) size is mm*n;
2-5, based on observation data X
NewSimilarity matrix S (X
New, θ), calculate new being divided into and cut parameter vector
Wherein || ||
2Expression second order norm, || ||
1Expression single order norm, α represents coefficient of dilatation, value is that 0.1, S ' be objective matrix S among the embodiment " one be listed as;
S " be the matrix that calculates according to Z,
2-6 judges whether the current iteration number of times reaches maximal value, as no current being divided into is cut parameter vector θ and is updated to new being divided into and cuts parameter vector
And return 3-2), otherwise parameter vector is cut in new being divided into
Be divided into as optimum and cut parameter vector.
By expectation maximization EM learning algorithm, obtain similarity matrix S according to the initial segmentation result X that obtains and initial partitioning parameters vector θ, calculating objective matrix S according to cutting apart the degree of confidence vector Z again, "; by iteration adjustment partitioning parameters vector θ so that similarity matrix S and objective matrix S " convergence is identical, is divided into and cuts parameter thereby obtain optimum.
Step 3, the characteristic model that generates is used for being divided into segmentation method based on the image of shortest path and notable feature, obtains segmentation result.The segmentation method that is divided into here is not limited to image based on shortest path and notable feature and is divided into and cuts, and those skilled in the art can easily expect processing this step with other existing segmentation method that is divided into.
Be illustrated in figure 1 as being divided into of different images group that obtains according to the present embodiment method and cut the result, can find out, based on the self-adaptation adjustment of image complexity analysis be divided into cut parameter to be divided into the segmentation result result that segmentation method obtains accurate, adaptive ability is strong.
Claims (6)
1. the feature adapting to image based on image complexity is divided into segmentation method, it is characterized in that, may further comprise the steps:
1) complexity of each image in the image sets of calculating input is chosen m the simple image of background in the image sets of input; M is the positive integer greater than 1;
2) m the simple image initial of background is partitioned into separately foreground area, obtains initial segmentation result;
3) utilizing initial segmentation result and initially being divided into to cut parameter obtains by the expectation maximization learning algorithm so that have the optimum of optimum similarity between the segmentation result of m the simple image of background and be divided into and cut parameter vector;
4) utilize optimum being divided into to cut parameter vector and all images in the input picture group are divided into to cut realize cutting apart shared object in all images.
2. a kind of feature adapting to image based on image complexity is divided into segmentation method as claimed in claim 1, it is characterized in that, the complexity of the image in the step 1) is specifically by the computational complexity mark
With the complexity mark
Embody:
Wherein,
For adopting multiple dimensioned image over-segmentation method based on the edge with image I
i, i=1 ..., the piece number of N over-segmentation under yardstick k, K is the maximum fractionation yardstick, the numbering of i correspondence image in image sets, image N is the sum of image in the image sets;
3. a kind of feature adapting to image based on image complexity is divided into segmentation method as claimed in claim 2, it is characterized in that,
The concrete grammar of choosing m the simple image of background in the step 1) in the image sets of input is:
Based on image I
iCorresponding complexity mark
Size each image is sorted, obtain ranking results, distribute image I
iSequence number after the corresponding ordering
Based on image I
iCorresponding complexity mark
Size each image is sorted, obtain ranking results, distribute image I
iSequence number after the corresponding ordering
Image I
iCorresponding final ranking results is
Final ranking results taking-up complexity according to each image in the image sets is low that front m image is the simple image of background.
4. a kind of feature adapting to image based on image complexity is divided into segmentation method as claimed in claim 1, it is characterized in that,
Step 2) by the Object Segmentation method based on remarkable detection m the simple image initial of background cut apart in.
5. a kind of feature adapting to image based on image complexity is divided into segmentation method as claimed in claim 1, it is characterized in that,
Step 3) specifically comprises following substep:
3-1) read default being divided into and cut parameter vector;
3-2) cut parameter vector θ and current segmentation result is asked for current similarity matrix S (X, θ) according to current being divided into; S (X, θ)=X θ=(X
1, X
2..., X
n) θ, wherein, being divided into and cutting parameter vector θ is all characteristic parameter ω
1..., ω
nThe transposition θ of the row vector that forms=(ω
1..., ω
n)
T, n is the total number of feature,
P, q=1 ..., m, l=1 ..., n, m represent to cut apart quantity,
D () representation feature distance function,
Be image I
pL feature in the foreground area that is partitioned into,
Be image I
qL the feature that is partitioned into; Initially current is divided into and cuts parameter vector θ is that parameter vector is cut in default being divided into, and initial current segmentation result X is initial segmentation result; 3-3) ask for current degree of confidence vector Z: the Z=V/max (V) of cutting apart based on similarity matrix S (X, θ); Wherein, V={v
1, v
2..., v
mFor length is the column vector of m,
P=1 ..., m,
Expression among the similarity matrix S with image I
pAll relevant row of segmentation result sue for peace;
3-4) based on the current degree of confidence vector Z of cutting apart, ask for observation data X
New:
X
New(r, t)=X (r, t) S " " (r), 1≤r≤mm, 1≤t≤n, X (r, t) represents that the position is at the numerical value of (r, t) in the current segmentation result matrix, S " " (r)=min (Z (p), Z (q)), S " " (r) represents objective matrix S " delegation, the capable data of r are by two image I of correspondence
p, I
qThe degree of confidence of cutting apart obtain, Z (p) is for cutting apart the capable data of degree of confidence vector Z p, Z (q) is for cutting apart the capable data of degree of confidence vector Z q;
3-5) based on observation data X
NewSimilarity matrix S (X
New, θ), calculate new being divided into and cut parameter vector
3-6) judge whether the current iteration number of times reaches maximal value, as no current being divided into cut parameter vector and be updated to new being divided into and cut parameter vector
And return 3-2), otherwise parameter vector is cut in new being divided into
Be divided into as optimum and cut parameter vector.
6. a kind of feature adapting to image based on image complexity is divided into segmentation method as claimed in claim 1, it is characterized in that,
All images in the input picture group are divided into cut by the segmentation method that is divided into based on shortest path and notable feature in the step 5).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210448129.XA CN102982539B (en) | 2012-11-09 | 2012-11-09 | Characteristic self-adaption image common segmentation method based on image complexity |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210448129.XA CN102982539B (en) | 2012-11-09 | 2012-11-09 | Characteristic self-adaption image common segmentation method based on image complexity |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102982539A true CN102982539A (en) | 2013-03-20 |
| CN102982539B CN102982539B (en) | 2015-05-27 |
Family
ID=47856481
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210448129.XA Expired - Fee Related CN102982539B (en) | 2012-11-09 | 2012-11-09 | Characteristic self-adaption image common segmentation method based on image complexity |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102982539B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103268614A (en) * | 2013-05-31 | 2013-08-28 | 电子科技大学 | Generation method for prospect spectrogram for multi-prospect co-segmentation |
| CN103268607A (en) * | 2013-05-15 | 2013-08-28 | 电子科技大学 | Common object detection method on weak supervision condition |
| CN103413310A (en) * | 2013-08-15 | 2013-11-27 | 中国科学院深圳先进技术研究院 | Collaborative segmentation method and device |
| CN103606143A (en) * | 2013-09-11 | 2014-02-26 | 电子科技大学 | Co-segmentation result correction method based on segmentation propagation |
| CN104573671A (en) * | 2014-10-28 | 2015-04-29 | 清华大学 | Method for discovering subject targets from video sequence |
| CN105427322A (en) * | 2015-12-03 | 2016-03-23 | 中国科学院地理科学与资源研究所 | Automatic selection method of high resolution remote sensing image segmentation parameter |
| CN107835378A (en) * | 2017-11-27 | 2018-03-23 | 韩劝劝 | A kind of image shot by cell phone complex method |
| CN108038502A (en) * | 2017-12-08 | 2018-05-15 | 电子科技大学 | Object collaborative detection method based on convolutional neural networks |
| CN108460768A (en) * | 2018-01-29 | 2018-08-28 | 北京航空航天大学 | The video perpetual object dividing method and device of stratification time domain cutting |
| CN109242864A (en) * | 2018-09-18 | 2019-01-18 | 电子科技大学 | Image segmentation result quality evaluating method based on multiple-limb network |
| CN109509345A (en) * | 2017-09-15 | 2019-03-22 | 富士通株式会社 | Vehicle detection apparatus and method |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010067277A1 (en) * | 2008-12-09 | 2010-06-17 | Koninklijke Philips Electronics N.V. | Image segmentation |
| CN101763514A (en) * | 2010-01-15 | 2010-06-30 | 西安电子科技大学 | Image segmentation method based on characteristic importance sorting spectral clustering |
| WO2012041419A1 (en) * | 2010-10-01 | 2012-04-05 | Telefonica, S.A. | Method and system for images foreground segmentation in real-time |
| CN102609903A (en) * | 2011-12-21 | 2012-07-25 | 辽宁师范大学 | Method for segmenting moveable outline model image based on edge flow |
| CN102722883A (en) * | 2012-04-16 | 2012-10-10 | 上海交通大学 | Polarized SAR (synthetic aperture radar) image segmentation method with space adaptivity |
| CN102750700A (en) * | 2012-06-05 | 2012-10-24 | 南京师范大学 | Fast robust fuzzy C-means image segmentation method combining neighborhood information |
-
2012
- 2012-11-09 CN CN201210448129.XA patent/CN102982539B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010067277A1 (en) * | 2008-12-09 | 2010-06-17 | Koninklijke Philips Electronics N.V. | Image segmentation |
| CN101763514A (en) * | 2010-01-15 | 2010-06-30 | 西安电子科技大学 | Image segmentation method based on characteristic importance sorting spectral clustering |
| WO2012041419A1 (en) * | 2010-10-01 | 2012-04-05 | Telefonica, S.A. | Method and system for images foreground segmentation in real-time |
| CN102609903A (en) * | 2011-12-21 | 2012-07-25 | 辽宁师范大学 | Method for segmenting moveable outline model image based on edge flow |
| CN102722883A (en) * | 2012-04-16 | 2012-10-10 | 上海交通大学 | Polarized SAR (synthetic aperture radar) image segmentation method with space adaptivity |
| CN102750700A (en) * | 2012-06-05 | 2012-10-24 | 南京师范大学 | Fast robust fuzzy C-means image segmentation method combining neighborhood information |
Non-Patent Citations (4)
| Title |
|---|
| BOGDAN ALEXE等: "What is an object?", 《IN PROC. IEEE CONF. COMPUT. VIS. PATTERN RECOGNIT.》, 30 June 2010 (2010-06-30), pages 73 - 80 * |
| MING-MING CHENG等: "Global Contrast based Salient Region Detection", 《IN PROC. IEEE CONF. COMPUT. VIS. PATTERN RECOGNIT.》, 30 June 2011 (2011-06-30), pages 409 - 416 * |
| PABLO ARBELAEZ等: "From Contours to Regions: An Empirical Evaluation", 《IN PROC. IEEE CONF. COMPUT. VIS.》, 30 June 2009 (2009-06-30), pages 2294 - 2301 * |
| SEUNG-JEAN KIM等: "An Interior-Point Method for Large-Scale l1-Regularized Least Squares", 《IEEE J. SEL. TOPICS SIGNAL PROCESS》, vol. 1, no. 4, 30 October 2007 (2007-10-30), pages 606 - 617 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103268607A (en) * | 2013-05-15 | 2013-08-28 | 电子科技大学 | Common object detection method on weak supervision condition |
| CN103268607B (en) * | 2013-05-15 | 2016-10-12 | 电子科技大学 | A kind of common object detection method under weak supervision condition |
| CN103268614A (en) * | 2013-05-31 | 2013-08-28 | 电子科技大学 | Generation method for prospect spectrogram for multi-prospect co-segmentation |
| CN103268614B (en) * | 2013-05-31 | 2016-01-20 | 电子科技大学 | A kind of for many prospects be divided into cut prospect spectrum drawing generating method |
| CN103413310B (en) * | 2013-08-15 | 2016-09-07 | 中国科学院深圳先进技术研究院 | Collaborative dividing method and device |
| CN103413310A (en) * | 2013-08-15 | 2013-11-27 | 中国科学院深圳先进技术研究院 | Collaborative segmentation method and device |
| CN103606143A (en) * | 2013-09-11 | 2014-02-26 | 电子科技大学 | Co-segmentation result correction method based on segmentation propagation |
| CN103606143B (en) * | 2013-09-11 | 2016-05-18 | 电子科技大学 | A kind ofly cut modified result method based on cutting apart being divided into of propagation |
| CN104573671A (en) * | 2014-10-28 | 2015-04-29 | 清华大学 | Method for discovering subject targets from video sequence |
| CN104573671B (en) * | 2014-10-28 | 2018-02-02 | 清华大学 | One kind finds theme mesh calibration method from video sequence |
| CN105427322A (en) * | 2015-12-03 | 2016-03-23 | 中国科学院地理科学与资源研究所 | Automatic selection method of high resolution remote sensing image segmentation parameter |
| CN105427322B (en) * | 2015-12-03 | 2018-02-13 | 中国科学院地理科学与资源研究所 | A kind of high score Remote Sensing Image Segmentation parameter automatic selecting method |
| CN109509345A (en) * | 2017-09-15 | 2019-03-22 | 富士通株式会社 | Vehicle detection apparatus and method |
| CN107835378A (en) * | 2017-11-27 | 2018-03-23 | 韩劝劝 | A kind of image shot by cell phone complex method |
| CN108038502A (en) * | 2017-12-08 | 2018-05-15 | 电子科技大学 | Object collaborative detection method based on convolutional neural networks |
| CN108460768A (en) * | 2018-01-29 | 2018-08-28 | 北京航空航天大学 | The video perpetual object dividing method and device of stratification time domain cutting |
| CN109242864A (en) * | 2018-09-18 | 2019-01-18 | 电子科技大学 | Image segmentation result quality evaluating method based on multiple-limb network |
| CN109242864B (en) * | 2018-09-18 | 2021-09-24 | 电子科技大学 | Image segmentation result quality evaluation method based on multi-branch network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102982539B (en) | 2015-05-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102982539B (en) | Characteristic self-adaption image common segmentation method based on image complexity | |
| CN105528589B (en) | Single image crowd's counting algorithm based on multiple row convolutional neural networks | |
| CN108648161B (en) | Binocular vision obstacle detection system and method of asymmetric kernel convolution neural network | |
| CN111832655B (en) | Multi-scale three-dimensional target detection method based on characteristic pyramid network | |
| CN103413151B (en) | Hyperspectral image classification method based on figure canonical low-rank representation Dimensionality Reduction | |
| CN104408733B (en) | Object random walk-based visual saliency detection method and system for remote sensing image | |
| CN105913087A (en) | Object identification method based on optimal pooled convolutional neural network | |
| CN109948593A (en) | Based on the MCNN people counting method for combining global density feature | |
| CN110781829A (en) | Light-weight deep learning intelligent business hall face recognition method | |
| CN104992403B (en) | Hybrid operator image redirection method based on visual similarity measurement | |
| CN104751175B (en) | SAR image multiclass mark scene classification method based on Incremental support vector machine | |
| CN103177450A (en) | Image scene segmentation and layering joint solution method based on component set sampling | |
| CN108197669B (en) | Feature training method and device of convolutional neural network | |
| CN104732545A (en) | Texture image segmentation method combined with sparse neighbor propagation and rapid spectral clustering | |
| CN102902956A (en) | Ground-based visible cloud image recognition processing method | |
| CN106355607B (en) | A kind of width baseline color image template matching method | |
| CN111339818A (en) | Face multi-attribute recognition system | |
| CN106650744A (en) | Image object co-segmentation method guided by local shape migration | |
| CN106910202A (en) | The image partition method and system of a kind of remote sensing images atural object | |
| CN107909079A (en) | One kind collaboration conspicuousness detection method | |
| CN109784401A (en) | A kind of Classification of Polarimetric SAR Image method based on ACGAN | |
| CN111488794A (en) | Adaptive receptive wildman population density estimation method based on hole convolution | |
| CN111414954A (en) | Rock image retrieval method and system | |
| CN105205816A (en) | Method for extracting high-resolution SAR image building zone through multi-feature weighted fusion | |
| CN102074013A (en) | Wavelet multi-scale Markov network model-based image segmentation method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150527 Termination date: 20171109 |