CN1570973A - An image retrieval method using marked edge - Google Patents
An image retrieval method using marked edge Download PDFInfo
- Publication number
- CN1570973A CN1570973A CN 03134425 CN03134425A CN1570973A CN 1570973 A CN1570973 A CN 1570973A CN 03134425 CN03134425 CN 03134425 CN 03134425 A CN03134425 A CN 03134425A CN 1570973 A CN1570973 A CN 1570973A
- Authority
- CN
- China
- Prior art keywords
- edge
- image
- prominent edge
- prominent
- carry out
- 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
Abstract
This invention is about a image search method which used of distinctive edge. It applied for the image that has relatively clear-cut edge. The basic point is selecting the edge that reflects the image character and using the distinctive edge. The characteristic is: detecting and refining edge of searched image, obtaining the outline image. Then by using independent edge self-intensify method, it picks the distinctive edge of image through repeated random heuristics search and intensifies for edge point. Then describing each distinctive edge by three typical features: cross-point rate, rotation frequency and corner-point rate and generating the character vector of image. If match, it measures the similarity degree between images by the method of overall distinctive edge matching. At last, export the similar image set according to the sequencing of similarity degree.
Description
Affiliated technical field: the present invention relates to a kind of method of utilizing prominent edge to carry out image retrieval, belong to fields such as computer vision, image understanding and pattern-recognition.Be applicable to relatively distinct image of edge, the graphical design that the retrieval edge obtains easily.
Background technology: since the nineties, along with computer technology, multimedia technology and rapid development of network technology, increasing image appears in the daily life.The feasible management and retrieval to image of the explosive increase of view data become key.At present, many image search methods all use shape to describe the feature of image.Shape is one of essential characteristic that characterizes on object, and under many circumstances, people often just can recognition object only according to the shape information of object, and this is the key point that shape is different from other visual signature such as color, texture etc.The shape description method that is applied at present in the image retrieval roughly can be divided into two classes: based on the image border with based on the zone.Use the marginal information of object to describe and query image based on the method for image border, these class methods are applicable to that the image border is comparatively clear, the more or less freely image that obtains.Mainly rely on the color distribution information of pixel in the zone to describe image based on the method in zone, these class methods for the zone can split comparatively accurately, in the zone color distribution comparatively the image of homogeneous is comparatively suitable.
Researchist and technician have proposed multiple image search method based on the image border, but still have a lot of problems.Method 1: at first extract marginal information, each bar edge all uses a polygon to be similar to, and the shape information of representative image is come on the summit of approximate polygon.The shortcoming of the method is: it requires the boundary curve closure, and this requirement is difficult to reach for general pattern.Method 2: a kind of elasticity of shape matching algorithm carries out image retrieval, at first by artificial appointment area-of-interest, adopts the optimized Algorithm of climbing the mountain to obtain the image border in these zones, uses the edge representative shape in these area-of-interests.The advantage of this method is that the edge is screened, and shortcoming is to need manual intervention, and this is often unrealistic in image retrieval.Method 3: use sketch to carry out image retrieval, in their work, image at first passes through a series of processing such as yardstick normalization, rim detection, refinement, simply the edge image that obtains and user's sketch is mated according to template then.Method 4: adopt the corner point on the edge to describe shape, one, on the basis of flex point, carry out affined transformation, utilize affine invariant features to represent shape.Two, adopt the phase place histogram quarter picture shape feature of marginal point.Three, propose a kind of curvature scale space method and describe shape, it is level and smooth for the Gauss that each bar boundary curve carries out under the different scale, under each yardstick, extract the bigger point of curvature, be chosen at the multiple dimensioned the longest point of life span down and describe boundary curve.The common drawback of above-mentioned several method is: only considered the information of special marginal point, and these marginal points can not be portrayed the shape of object well.Method 5: a kind of architectural feature based on the edge is described shape, and it uses and " pours water
i± (Water-Filling) algorithm extracts boundary curve, each bar boundary curve with some architectural features as: irrigation period, crunode number, crunode histogram wait and represent, and the shape facility of entire image with several " special " boundary curve as: the maximum edges of crunode, the longest curve of irrigation period wait quarter.The advantage of this method is: utilized the structural information of boundary curve but not single marginal points information is represented shape, its shortcoming is: only use other boundary curve, often because the error of noise or extraction edge process is and inaccurate, this will reduce the accuracy rate of retrieval to these boundary curves.Method 6: a kind of method based on Fourier analysis, this method at first obtains the fundamental function that can describe shape, as based on the fundamental function of curvature or based on the fundamental function of radius.Then this fundamental function is made discrete Fourier transform (DFT), the fourier coefficient after the use conversion comes retrieving images as shape facility.This method is quite responsive to the subtle change and the noise of edge point position, therefore, can not obtain good effect in actual retrieval.Method 7: tessellation, at first choose angle point on the image border as unique point, use the Delaunay triangle to divide then, can write down leg-of-mutton shape facility and describe feature of image shape.This method is because also based on some particular points on the edge, so also comparatively responsive for the variation of noise and some position.
The method that present existing dependence marginal information is described picture shape and then retrieving images has following two main defectives: one, before utilizing the edge extracting shape information edge is not analyzed and selected, most of algorithms have used all edges in the image.The purpose of image retrieval is in order to search out similar image, and in actual treatment, because the inaccuracy and the The noise of edge extracting, not all edge all can produce positive role to describing picture shape and images match; Two, during the similarity between the dimensioned plan picture, adopted simple " one to one " matching strategy.This matching strategy calculates quite simple, but because the inaccurate of The noise and edge extracting tends to make the longest irrigation period edge that extracts with the crunode edge is inaccurate at most, directly causes the appearance of mistake coupling, thus the accuracy rate that influence is retrieved.
Summary of the invention: for avoiding the defective of prior art, the present invention is at the describing method of systematically having studied based on the shape facility at edge, and the application of shape facility in image retrieval, has proposed a kind of image search method based on prominent edge.Different with other method, we think can represent prominent edge in the image that shape should be, and the algorithm that has designed an independent edges self-enhancement extracts the prominent edge in the image.Then, we use three features to describe each bar prominent edge, and then form the eigenvector of image.During similarity between the dimensioned plan picture, we do not adopt traditional " one to one " matching criterior, but have used the matching criterior of a kind of " multi-to-multi ", its objective is in order to reduce because the inaccurate harmful effect that retrieval is caused of Edge extraction.A large amount of experiment showed, that method proposed by the invention with respect to other method, has good performance: 1, owing to used prominent edge, reject short and small edge, removed a unfavorable factor that influences retrieval rate.Also reduce simultaneously calculated amount, improve retrieval rate; 2, adopt the matching strategy of " multi-to-multi ", can reduce the inaccurate influence in edge to a certain extent.
Basic thought of the present invention is that the edge of describing characteristics of image is allowed a choice, and adopts prominent edge.They all have certain robustness for scale, noise, edge extracting are inaccurate etc.Prominent edge be meant that visual intensity in the image is big and length is long the edge, it is characterized in that: at first query image is carried out rim detection and refinement, obtain outline map; Secondly, use the separate edge self-strengthening method,, pick out the prominent edge in the image by to marginal point heuristic search immediately and enhancing repeatedly; Then, use three characteristic features for each bar prominent edge, crunode rate, rotational frequency and angle point rate are described, and then generate the eigenvector of image; Then, the time adopt similarity degree between the method dimensioned plan picture of comprehensive prominent edge coupling in coupling; At last, according to sequencing of similarity, the set of output similar image.
The separate edge self-enhancement is that marginal point is independently connected into boundary curve, and the boundary curve that obtains is carried out suitable processing, for follow-up prominent edge is selected to facilitate.This method is: the strength information with the edge is measured as guiding, heuristic search at random by repeatedly obtains various possible image borders, utilize integrator that each bar separate edge is carried out self-enhancement then, so just make the prominent edge in the image obtain very big enhancing, utilize the result after strengthening just to choose prominent edge easily at last.The great advantage of independent edges self-strengthening method is: the amplitude that the edge strengthens is directly proportional with the significance degree of self, and therefore, the result after the enhancing more helps the selection of prominent edge.In addition, because heuristic search at random repeatedly makes that the process of extracting the edge is affected by noise less.
Because the Canny operator has good location and detailed performance, so the rim detection among the present invention adopts the Canny operator.Image after the rim detection is referred to as outline map (Edge map), and its edge strength has been represented in the brightness of every bit among the figure, and the big more edge strength that means of brightness is big more, can find the picture element that gray scale is undergone mutation in the part.
The generation of characteristics of image vector: for each bar prominent edge, adopt three characteristic features, crunode rate, rotational frequency and angle point rate are described.
The crunode rate: the bifurcation number of boundary curve can be weighed the complex structure degree at edge well.Each bar prominent edge is corresponded in the original outline map, along the end points of prominent edge, carry out " pouring water ", the definition bifurcated is counted and is the total degree of the time-division fork that flows along the edge when current.Then the crunode rate of this prominent edge is: the total degree of bifurcated/its length.For prominent edge, high more its structure of explanation of its crunode rate is complicated more.
Rotational frequency: the rotational frequency of curve is the degree of crook that is used for describing the edge.The rotational frequency of each bar prominent edge is: the total degree that this edge rotates/its length.This boundary curve degree of crook of the high more expression of rotational frequency is big more.
The angle point rate: the corner point frequency is used for weighing the level and smooth degree of boundary curve.Corner point is the important particular point of a class on the boundary curve, and the corner point bright edge trend of speaking more more is strong at localized variation Shaoxing opera, and the edge is unsmooth more on the whole.The angle point rate of each bar prominent edge is: corner point number/its length.
Be defined in the segment of curve among a small circle, can only have a corner point.
After having determined the primitive character of each bar prominent edge, just obtain the prominent edge set of an image, also just obtain the characteristics of image vector.
Owing to adopt crunode rate, rotational frequency and angle point rate as the primitive character of describing image, portrayed curvilinear characteristic from complex structure degree, degree of crook peace slippage degree three aspects of curve, calculate quite simple.These three primitive characters are all insensitive for translation and rotation, promptly satisfy translation and rotational invariance.As for the yardstick unchangeability, this is quite inappeasable for the eigenvector that is used for carrying out image retrieval, because our three primitive characters have all used ratio, so can satisfy that shape description is should the picture engraving feature accurate, algorithm is simple, easy to operate, also should have this character of unchangeability to rotation, translation, scale.
When images match, the similarity between the image is generally measured by the distance between the character pair vector, and is final, and the minimum image collection of distance is thought similar image.The comprehensive prominent edge matching process that the present invention proposes about images match does not adopt traditional " one to one " coupling, but loosened for the strict demand of mating, it has adopted the matching scheme of a kind of " multi-to-multi ", a prominent edge of piece image allows to be complementary with many prominent edges of another width of cloth image, concrete matching strategy relies on two criterions to retrain, that is: importance degree satisfies criterion: with the most similar matching criterior at first.Similarity final between image is decided by all effective couplings.The outstanding advantage of the method is: the mistake that has reduced to cause because of edge extracting is inaccurate is mated, and can improve retrieval rate to a certain extent.
Description of drawings:
Fig. 1: the basic flow sheet of the inventive method
Fig. 2: with system give an example the inquiry example
Fig. 3: with system give an example the inquiry example
(a) sketch of a width of cloth user Freehandhand-drawing
(b) according to the result of user's freehand sketch retrieval
Embodiment:
Now in conjunction with the accompanying drawings the present invention is further described:
The image search method based on prominent edge that proposes according to the present invention, we have realized the prototype system of an image retrieval with C Plus Plus.At present, have 4500 width of cloth images in our image data base, these images comprise: buildings, landscape, trade mark, icon, people's face etc.The source of image has: online download, Corel stock photo library extract, digital camera is taken and Yale research Lab face database.Image in our image data base is gray level image and the edge is all comparatively clear.
Suppose an image Q to be checked, the retrieval image I similar, that is: D (X in image data base to Q
Q, X
I)≤t.D is the distance function of eigenvector in the following formula, and t is the threshold value of being set up by the user, X
QBe the eigenvector of image Q, X
IIt is the eigenvector of image I.The similarity degree of two width of cloth image Q and I can be with their eigenvector X
QAnd X
IDistance represent that distance more little expression two images are similar more.The result of inquiry changes along with threshold value, and the satisfied all the time distance with image to be checked is less than or equal to threshold value.The user can be also directly to require system's output and the most similar image collection of image to be checked, as output and 20 nearest width of cloth images of image distance to be checked.
At first image Q is carried out rim detection and refinement, obtain outline map.Use the separate edge self-strengthening method then: promptly, pick out the prominent edge in the image by to marginal point heuristic search immediately and enhancing repeatedly.Calculate crunode rate, rotational frequency and the angle point rate of each bar prominent edge.Suppose a prominent edge C
i, its length is l
i, the bifurcation number that is gone out by " pouring water " algorithm computation is f
CiC then
iCrunode rate fr
i=f
Ci/ l
iRotational frequency rf
C=rn
T/ l
i, rn wherein
TBe edge prominent edge C
iThe total degree that rotates; When its corner point number is cn
iThe time, flex point rate cf then
i=cn
i/ l
i
After having determined the primitive character of each bar prominent edge, we just can obtain the eigenvector of image.Prominent edge set C={c according to image Q
1, c
2..., c
i..., c
V,, use f
1, f
2..., f
i..., f
VRepresent prominent edge c respectively
1, c
2..., c
i..., c
VFeature, then have:
f
1＝(fr
1，rf
1，cf
1)，f
2＝(fr
2，rf
2，cf
3)，...，f
i(fr
i，rf
i，cf
i，...，f
L＝(fr
V，rf
V，cf
V)
So, the eigenvector of image Q
For:
Carry out images match, adopt comprehensive prominent edge matching process, prominent edge of piece image allows the matching scheme of " multi-to-multi " that many prominent edges with another width of cloth image are complementary to carry out images match.The eigenvector of image Q
For:
The eigenvector f of image Q ' so
Q' be: f
Q'=[f
1', f
2' ..., f
j' ..., f
V] '.Measure with the distance between the character pair vector, final, the minimum image collection of distance is thought similar image.
The support user of system that the present invention implemented carries out two kinds of inquiries: inquiry (Query by example) and sketch inquiry (Query by sketch) for example.Inquiry is meant by the user image to be checked is provided for example, and by the automatic output of system some width of cloth images similarly, the number of similar image can be specified by the user, and scope is between 0 to 100.Sketch inquiry is meant by user's width of cloth sketch that draws submits to system queries, and similarly, the user can appointing system exports the similar image of some numbers.In query script, the user can check its relevant information such as size, source by the output image of " double-click " system, the user also can with a certain width of cloth Query Result image as an example image carry out new inquiry.
Fig. 2 has provided this chapter searching system Query Result example preferably of giving an example, and first width of cloth image in the image upper left corner is a query image, beats the correct result for retrieval of expression of " √ ", and beats the result for retrieval of the expression mistake of " * ".Fig. 3 has provided this chapter searching system and has carried out sketch Query Result example preferably, and (a) the middle image that shows is the sketch of user's Freehandhand-drawing, and (b) image in is the result of retrieval.
Find out from experimental result: the prominent edge that this method is extracted relatively meets people's subjective judgement; Higher with respect to other method retrieval rate; Owing to used the matching strategy of prominent edge and " multi-to-multi ", therefore had the higher search accuracy rate.
Claims (8)
1, a kind of method of utilizing prominent edge to carry out image retrieval is characterized in that: at first query image is carried out rim detection and refinement, obtain outline map; Secondly, use the separate edge self-strengthening method,, pick out the prominent edge in the image by to marginal point heuristic search immediately and enhancing repeatedly; Then, use three characteristic features for each bar prominent edge, crunode rate, rotational frequency and angle point rate are described, and then generate the eigenvector of image; Then, the time adopt similarity degree between comprehensive prominent edge Matching Algorithm dimensioned plan picture in coupling; At last, according to sequencing of similarity, the set of output similar image.
2, a kind of method of utilizing prominent edge to carry out image retrieval according to claim 1, it is characterized in that: the separate edge self-strengthening method: the strength information with the edge is measured as guiding, heuristic search at random by repeatedly obtains various possible image borders, utilize integrator that each bar separate edge is carried out self-enhancement then, utilize the result after strengthening just to choose prominent edge easily at last.
3, a kind of method of utilizing prominent edge to carry out image retrieval according to claim 1, it is characterized in that: comprehensive prominent edge matching process, adopted the matching scheme of a kind of " multi-to-multi ", a prominent edge of piece image allows to be complementary with many edges of another width of cloth image; Concrete matching strategy relies on two criterions to retrain, that is: importance degree satisfies criterion and the most similar matching criterior at first, and similarity final between image is decided by all effective couplings.
4, a kind of method of utilizing prominent edge to carry out image retrieval according to claim 1 is characterized in that: rim detection adopts the Canny operator.
5, a kind of method of utilizing prominent edge to carry out image retrieval according to claim 1, it is characterized in that: each bar prominent edge is corresponded in the original outline map, end points along prominent edge, carry out " pouring water ", the definition bifurcated is counted and is that then the crunode rate of this prominent edge is when flow along the edge total degree of time-division fork of current: the total degree of bifurcated/its length.
6, a kind of method of utilizing prominent edge to carry out image retrieval according to claim 1, it is characterized in that: the rotational frequency of each bar prominent edge is: the total degree that this edge rotates/its length.This boundary curve degree of crook of the high more expression of rotational frequency is big more.
7, a kind of method of utilizing prominent edge to carry out image retrieval according to claim 1, it is characterized in that: the angle point rate of each bar prominent edge is: corner point number/its length.
8, a kind of method of utilizing prominent edge to carry out image retrieval according to claim 7 is characterized in that: be defined in the segment of curve among a small circle, can only have a corner point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 03134425 CN1290061C (en) | 2003-07-23 | 2003-07-23 | An image retrieval method using marked edge |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 03134425 CN1290061C (en) | 2003-07-23 | 2003-07-23 | An image retrieval method using marked edge |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1570973A true CN1570973A (en) | 2005-01-26 |
| CN1290061C CN1290061C (en) | 2006-12-13 |
Family
ID=34470195
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 03134425 Expired - Fee Related CN1290061C (en) | 2003-07-23 | 2003-07-23 | An image retrieval method using marked edge |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1290061C (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100589520C (en) * | 2007-09-14 | 2010-02-10 | 西北工业大学 | Detection method for characteristic of edge and angle point of color image |
| CN101287133B (en) * | 2007-09-14 | 2010-06-23 | 西北工业大学 | Digital symmetric oriented tensorial filtering method |
| CN101493936B (en) * | 2008-05-30 | 2011-03-23 | 内蒙古科技大学 | Multi- resolution non-rigid head medicine image registration method based on image edge |
| US8027550B2 (en) | 2007-03-30 | 2011-09-27 | Sharp Kabushiki Kaisha | Image-document retrieving apparatus, method of retrieving image document, program, and recording medium |
| CN102339306A (en) * | 2010-08-31 | 2012-02-01 | 微软公司 | Sketch-based image search |
| CN101621710B (en) * | 2009-07-21 | 2012-07-11 | 深圳市融创天下科技股份有限公司 | Method and system for evaluating video quality based on edge detection |
| CN102609911A (en) * | 2012-01-16 | 2012-07-25 | 北方工业大学 | Edge-based image significance detection |
| CN102902807A (en) * | 2011-10-18 | 2013-01-30 | 微软公司 | Visual search using a pluraligy of visual input modal |
| US8472665B2 (en) | 2007-05-04 | 2013-06-25 | Qualcomm Incorporated | Camera-based user input for compact devices |
| CN103999084A (en) * | 2011-12-27 | 2014-08-20 | 索尼公司 | Server, client terminal, system, and recording medium |
| CN104243821A (en) * | 2014-09-10 | 2014-12-24 | 广东欧珀移动通信有限公司 | Obtaining method and device of large-view-angle photo |
| US9507803B2 (en) | 2011-10-18 | 2016-11-29 | Microsoft Technology Licensing, Llc | Visual search using multiple visual input modalities |
| CN108733749A (en) * | 2018-04-08 | 2018-11-02 | 天津大学 | A kind of image search method based on sketch |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101799828B (en) * | 2010-03-11 | 2012-01-11 | 南昌航空大学 | Book lookup method based on perspective transformation for video point reading machine |
-
2003
- 2003-07-23 CN CN 03134425 patent/CN1290061C/en not_active Expired - Fee Related
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8027550B2 (en) | 2007-03-30 | 2011-09-27 | Sharp Kabushiki Kaisha | Image-document retrieving apparatus, method of retrieving image document, program, and recording medium |
| US8472665B2 (en) | 2007-05-04 | 2013-06-25 | Qualcomm Incorporated | Camera-based user input for compact devices |
| CN101287133B (en) * | 2007-09-14 | 2010-06-23 | 西北工业大学 | Digital symmetric oriented tensorial filtering method |
| CN100589520C (en) * | 2007-09-14 | 2010-02-10 | 西北工业大学 | Detection method for characteristic of edge and angle point of color image |
| CN101493936B (en) * | 2008-05-30 | 2011-03-23 | 内蒙古科技大学 | Multi- resolution non-rigid head medicine image registration method based on image edge |
| CN101621710B (en) * | 2009-07-21 | 2012-07-11 | 深圳市融创天下科技股份有限公司 | Method and system for evaluating video quality based on edge detection |
| CN102339306A (en) * | 2010-08-31 | 2012-02-01 | 微软公司 | Sketch-based image search |
| CN102902807A (en) * | 2011-10-18 | 2013-01-30 | 微软公司 | Visual search using a pluraligy of visual input modal |
| CN102902807B (en) * | 2011-10-18 | 2016-06-29 | 微软技术许可有限责任公司 | Use the visual search of multiple vision input mode |
| US9507803B2 (en) | 2011-10-18 | 2016-11-29 | Microsoft Technology Licensing, Llc | Visual search using multiple visual input modalities |
| CN103999084A (en) * | 2011-12-27 | 2014-08-20 | 索尼公司 | Server, client terminal, system, and recording medium |
| CN102609911A (en) * | 2012-01-16 | 2012-07-25 | 北方工业大学 | Edge-based image significance detection |
| CN102609911B (en) * | 2012-01-16 | 2015-04-15 | 北方工业大学 | Edge-based image significance detection |
| CN104243821A (en) * | 2014-09-10 | 2014-12-24 | 广东欧珀移动通信有限公司 | Obtaining method and device of large-view-angle photo |
| CN104243821B (en) * | 2014-09-10 | 2018-07-03 | 广东欧珀移动通信有限公司 | A kind of acquisition methods and device of big visual angle photo |
| CN108733749A (en) * | 2018-04-08 | 2018-11-02 | 天津大学 | A kind of image search method based on sketch |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1290061C (en) | 2006-12-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN1290061C (en) | An image retrieval method using marked edge | |
| Smith et al. | Measuring texture classification algorithms | |
| CN103077512B (en) | Based on the feature extracting and matching method of the digital picture that major component is analysed | |
| Wang et al. | Generalizing edge detection to contour detection for image segmentation | |
| Zou et al. | CrackTree: Automatic crack detection from pavement images | |
| CN108734694A (en) | Thyroid tumors ultrasonoscopy automatic identifying method based on faster r-cnn | |
| CN106778823A (en) | A kind of readings of pointer type meters automatic identifying method | |
| Chen et al. | A robust descriptor based on weber’s law | |
| CN112396612B (en) | Vector information assisted remote sensing image road information automatic extraction method | |
| CN108681737A (en) | A kind of complex illumination hypograph feature extracting method | |
| CN105023027A (en) | Sole trace pattern image retrieval method based on multi-feedback mechanism | |
| CN108268865A (en) | Licence plate recognition method and system under a kind of natural scene based on concatenated convolutional network | |
| CN109902585A (en) | A kind of three modality fusion recognition methods of finger based on graph model | |
| CN1286064C (en) | An image retrieval method based on marked interest point | |
| Rangkuti et al. | Batik image retrieval based on similarity of shape and texture characteristics | |
| CN110147780A (en) | The landform recognition methods of real-time field robot and system based on level landform | |
| Maaten et al. | Computer vision and machine learning for archaeology | |
| CN108182438A (en) | Figure binary feature learning method and device based on deeply study | |
| CN113658129B (en) | Position extraction method combining visual saliency and line segment strength | |
| CN113392856B (en) | Image forgery detection device and method | |
| US20050207653A1 (en) | Method for analysis of line objects | |
| CN103871084B (en) | Indigo printing fabric pattern recognition method | |
| CN115375891A (en) | Cultural relic fragment similarity identification and transformation matching method based on machine learning | |
| Liu et al. | An Image Segmentation Method for the blind sidewalks recognition by using the convolutional neural network U-net | |
| CN109299295A (en) | Indigo printing fabric image database search 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 | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20061213 |