CN103365850A - Method and device for annotating images - Google Patents
Method and device for annotating images Download PDFInfo
- Publication number
- CN103365850A CN103365850A CN2012100845545A CN201210084554A CN103365850A CN 103365850 A CN103365850 A CN 103365850A CN 2012100845545 A CN2012100845545 A CN 2012100845545A CN 201210084554 A CN201210084554 A CN 201210084554A CN 103365850 A CN103365850 A CN 103365850A
- Authority
- CN
- China
- Prior art keywords
- image
- label
- images
- described image
- collection
- 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 discloses a method and a device for annotating images. The method for annotating the images includes selecting a plurality of other images similar to each image in a given image set in the aspect of image features in the given image set; fitting the image features of each image by the aid of the image features of the various other images to acquire a plurality of fitting coefficients of the image; constructing a label of each image by the aid of labels of the various other images according to the fitting coefficients of the image.
Description
Technical field
Relate generally to image management of the present invention and image retrieval.Particularly, the present invention relates to a kind of method and apparatus that can be optimized the label of image.
Background technology
In recent years, along with developing rapidly of multimedia technology and internet, quantity and the complexity of digital picture increase rapidly.Therefore, exist and how great amount of images to be carried out fast and the effectively problem of management, relate generally to the access, access, tissue, retrieval of image etc.
Image is associated with some textual description information usually, and for example, title, descriptor, review information etc. are in order to the information such as content, spot for photography, personal feeling and evaluation that show image.Therefore, can be that image adds label based on these information, or directly with descriptor as label, come the management and retrieval of assistant images.For example, can directly use descriptor (label) to carry out image retrieval: when the user wishes to search some image, the user inputs keyword, the keyword of image retrieving apparatus comparison user input and existing descriptor (label), if comprise this keyword in the descriptor, so, think that this image is target image, and it is fed back to the user, thereby realized easily the image retrieval function based on keyword.Image retrieval based on keyword does not need the content of image is analyzed and compared, and is therefore, more simple, quick than traditional CBIR.In addition, can also effectively utilize some prior arts in the text retrieval.
Yet often there are some problems in above-mentioned textual description information.For instance, (1) accuracy is low, i.e. the descriptor content of Description Image exactly.For example, the textual description information of the photo of Zhangbei County capital Chinese courtyard house is " Pekinese's building ", although Chinese courtyard house also are a kind of buildings, and, usefulness " building " can not be described the content in the photo exactly.Find Chinese courtyard house but not the user of other building for hope, be difficult to utilize keyword " Chinese courtyard house " to find this width of cloth image.(2) incomplete, namely descriptor can not contain the main contents in the photo.For example the content of photo is the swan in the Qinghai Lake, and " swan " this label has just been described the partial content in the picture, for contents such as the trees of the lake in the picture, lakeside, skies, does not all have corresponding label.(3) ambiguity is arranged, namely given descriptor (label) versatility of user is not strong.For example the descriptor of the photo of a pet dog (label) may be the name of this pet dog, and this descriptor (label) does not just have versatility concerning other people.Above-mentioned and other problems has limited textual description information such as directly utilizing descriptor image has been managed effectively, therefore, need to carry out to the label of image necessary correction and replenishes.In addition, for the image that does not have label, need to add suitable label for it.
Summary of the invention
Provided hereinafter about brief overview of the present invention, in order to basic comprehension about some aspect of the present invention is provided.Should be appreciated that this general introduction is not about exhaustive general introduction of the present invention.It is not that intention is determined key of the present invention or pith, neither be intended to limit scope of the present invention.Its purpose only is that the form of simplifying provides some concept, with this as the in greater detail preorder of discussing after a while.
The objective of the invention is the problems referred to above for prior art, proposed a kind of method and apparatus that can be optimized the label of image.This scheme can be optimized the existing label of image quickly and accurately and be not have the image of label to add suitable label.
To achieve these goals, according to an aspect of the present invention, a kind of image labeling method is provided, has comprised: for each image in the given image collection, in described image collection, be chosen in a plurality of other images close with described image on the characteristics of image; By the characteristics of image with the described image of characteristics of image match of described a plurality of other images, obtain a plurality of fitting coefficients of described image; And according to described a plurality of fitting coefficients of described image, utilize the label of described a plurality of other images to construct the label of described image.
According to a specific embodiment of the present invention, the a plurality of fitting coefficients that obtain described image further comprise: by so that be used on the characteristics of image with Given Graph and come the error of match Given Graph picture minimum as close a plurality of other images, obtain described a plurality of fitting coefficients of described image.
According to a specific embodiment of the present invention, construct the label of described image to satisfy predetermined constraint condition.
According to a specific embodiment of the present invention, described constraint condition comprises that the label configurations total error of whole image collection is minimum.
According to a specific embodiment of the present invention, described constraint condition is relevant with the correlativity between the label.
According to a specific embodiment of the present invention, described constraint condition is relevant with original label of described image.
According to a specific embodiment of the present invention, the label of constructing described image further comprises: at random or an image in the described image collection of select progressively; Utilization comes the label of the selected image of match corresponding to the label of a plurality of other images of selected image with the fitting coefficient of correspondence; And the repetition above-mentioned steps, until be in the described image collection each image configuration label.
According to another aspect of the present invention, a kind of image labeling device is provided, comprise: neighbour's image collection module, be used for each image for given image collection, in described image collection, be chosen in a plurality of other images close with described image on the characteristics of image; The fitting coefficient acquisition module for the characteristics of image that passes through with the described image of characteristics of image match of described a plurality of other images, obtains a plurality of fitting coefficients of described image; And the image tag constructing module, be used for the described a plurality of fitting coefficients according to described image, utilize the label of described a plurality of other images to construct the label of described image.
According to a specific embodiment of the present invention, described image tag constructing module is constructed the label of described image to satisfy predetermined constraint condition.
According to a specific embodiment of the present invention, described constraint condition comprises that the label configurations total error of whole image collection is minimum.
According to a specific embodiment of the present invention, described constraint condition is relevant with the correlativity between the label.
According to a specific embodiment of the present invention, described constraint condition is relevant with original label of described image.
In addition, according to a further aspect in the invention, also provide a kind of storage medium.Described storage medium comprises machine-readable program code, and when when messaging device is carried out described program code, described program code is so that described messaging device executive basis said method of the present invention.
In addition, in accordance with a further aspect of the present invention, also provide a kind of program product.Described program product comprises the executable instruction of machine, and when when messaging device is carried out described instruction, described instruction is so that described messaging device executive basis said method of the present invention.
Description of drawings
With reference to below in conjunction with the explanation of accompanying drawing to the embodiment of the invention, can understand more easily above and other purpose of the present invention, characteristics and advantage.Parts in the accompanying drawing are just in order to illustrate principle of the present invention.In the accompanying drawings, same or similar technical characterictic or parts will adopt identical or similar Reference numeral to represent.In the accompanying drawing:
Fig. 1 shows the process flow diagram according to the image labeling method of the embodiment of the invention;
Fig. 2 shows the block diagram according to the image labeling device of the embodiment of the invention; And
Fig. 3 shows the schematic block diagram that can be used for implementing according to the computing machine of the method and apparatus of the embodiment of the invention.
Embodiment
In connection with accompanying drawing example embodiment of the present invention is described in detail hereinafter.For clarity and conciseness, all features of actual embodiment are not described in instructions.Yet, should understand, in the process of any this practical embodiments of exploitation, must make a lot of decisions specific to embodiment, in order to realize developer's objectives, for example, meet those restrictive conditions with system and traffic aided, and these restrictive conditions may change to some extent along with the difference of embodiment.In addition, might be very complicated and time-consuming although will also be appreciated that development, concerning the those skilled in the art that have benefited from present disclosure, this development only is routine task.
At this, what also need to illustrate a bit is, for fear of having blured the present invention because of unnecessary details, only show in the accompanying drawings with according to the closely-related apparatus structure of the solution of the present invention and/or treatment step, and omitted other details little with relation of the present invention.In addition, it is pointed out that also element and the feature described can combine with element and the feature shown in one or more other accompanying drawing or the embodiment in an accompanying drawing of the present invention or a kind of embodiment.
The present invention is based on following thought: if treat an image and label thereof isolatedly, label may be inaccurate.Yet if process all image-label datas from the viewpoint of statistics, so, it is correct that label is greatly arranged.Therefore, can optimize or construct by means of the label of other images the label of pending image.In addition, correct label always has correspondence with picture material, and the label of mistake then presents characteristic at random.Therefore, can from a large amount of image-label datas, extract the image-label pair of " having identical correspondence ", to realize the optimization to label.
The below describes the according to an embodiment of the invention flow process of image labeling method with reference to Fig. 1.
Fig. 1 shows the process flow diagram according to the image labeling method of the embodiment of the invention.As shown in Figure 1, according to image labeling method of the present invention, comprise the steps: for each image in the given image collection, in described image collection, be chosen in a plurality of other images (step S1) close with described image on the characteristics of image; By the characteristics of image with the described image of characteristics of image match of described a plurality of other images, obtain a plurality of fitting coefficients (step S2) of described image; And according to described a plurality of fitting coefficients of described image, utilize the label of described a plurality of other images to construct the label (step S3) of described image.
Label as herein described is expressed with vector form, and namely an image is corresponding to a label vector.For example, the content of an image is the tiger in the zoo, and its existing descriptor is zoo, tiger.Suppose that the label that the label vector comprises is zoo, tiger, lion, frog, cage, visitor.Then the existing label vector of this image is [1,1,0,0,0,0], is [1/2,1/2,0,0,0,0] after the normalization.
The each basis of processing of the present invention is a given image collection, for example, and the part or all of image in the image data base.It should be noted that the image in the given image collection must all not have label, as long as the parts of images in the given image collection has label, for the image that does not have label, its initial labels vector can be null vector.
At step S1, for each image in the given image collection, in described image collection, be chosen in a plurality of other images close with described image on the characteristics of image.
As mentioned above, in image space, each image can be similar to by its k arest neighbors image, and wishes that the label of this image also can be similar to by the label of this k arest neighbors image, and wherein k is positive integer.Therefore, for the label to image is optimized, need at first search with it at approximate a plurality of other images of characteristics of image (having reflected picture material).Like this, could then remove to optimize the label of pending image with the label of a plurality of other images.
It should be noted that may be different for k arest neighbors image of each image.In addition, for each image, the value of k also can be different.For example, for an image, search with its on characteristics of image recently like 3 images, and for another image, search with its on feature recently like 15 images.K value itself can be but must not be the criterion of search arest neighbors image.Can select other criterion, for example, choose all images that the similarity that satisfies characteristics of image surpasses predetermined threshold, the number of these images is the value of k.
Only for example, step S1 can realize by following step: at first, extract each Characteristic of Image in the given image collection; Then, calculate the distance (having reflected similarity) of the feature of Given Graph picture and residual image; At last, (being the similarity maximum) front k image of chosen distance minimum is as k arest neighbors image of Given Graph picture.Herein, choosing with calculating of characteristics of image and distance can be adopted method of the prior art.For example, can select color histogram feature, texture or shape facility, Euclidean distance etc.
It should be noted that in the present invention, for each image in the given image collection, all need to calculate its k arest neighbors image.As mentioned above, for each image, the k value can be different.
The below briefly introduces step S2.
At step S2, by the characteristics of image with the described image of characteristics of image match of described a plurality of other images, obtain a plurality of fitting coefficients of described image.That is, by the characteristics of image with the characteristics of image match present image of k arest neighbors image, obtain a plurality of fitting coefficients of embodiment image Relations Among, with among the step S3 below, utilize a plurality of fitting coefficients, the label of construct image.Therefore, the effect of step S2 mainly is to obtain fitting coefficient, i.e. relation between Given Graph picture and its k the arest neighbors image.
Therefore, at step S2, for each Given Graph picture and k arest neighbors image thereof, be fitting coefficient of each generation of k arest neighbors image, namely obtain the fitting coefficient vector of each Given Graph picture.
The below is take a Given Graph picture and k arest neighbors image thereof as example.
Suppose that the present image characteristic of correspondence is x
i, its k arest neighbors Characteristic of Image is
The fitting coefficient vector is W={w
1..., w
k.
At first, calculating size is the correlation matrix C of k * k, and m element capable, the n row is in this matrix:
m，n＝1，....，k。
Then, separate linear system C*W=1, obtain fitting coefficient vector W.Find the solution above-mentioned linear equations and can adopt existing method.
At last, with each coefficient normalization of fitting coefficient vector W.The value that is about to each element among the fitting coefficient vector W divided by all these elements and.
Also can adopt other method that can obtain according to characteristics of image the corresponding a plurality of fitting coefficients of each image herein.
The below specifically describes step S3.
At step S3, according to described a plurality of fitting coefficients of described image, utilize the label of described a plurality of other images to construct the label of described image.
The below illustrates a kind of specific implementation of step S3-recurrence revised law.
At first, initialization label vector.Be the label vector of each image setting in the given image collection, this vectorial length is total number of labels order M, and each element in this vector is corresponding to a label.The value of its label vector is set according to the initial labels of each image.If an image has certain initial labels, then corresponding element is set to 1 in its label vector, otherwise, be set to 0.If an image does not have initial labels, then the value of each element is 0 or the value rule of thumb set in its label vector.Then, normalization label vector.
Then, at random or an image in the described image collection of select progressively.Utilization comes the label of the selected image of match corresponding to the label of a plurality of other images of selected image with the fitting coefficient of correspondence.Repeat above-mentioned steps, until be in the described image collection each image configuration label.
If the label vector of the k of this image arest neighbors image is respectively Y
i, i=1 ..., k, the fitting coefficient of this k image is W={w
1..., w
k, then the new label vector of this image is:
Therefore, can repeat above-mentioned steps and reach predetermined number of times, or repeat above-mentioned steps until the label vector of all images no longer change or change less till, coming is that the new label of each image configuration in the image collection is vectorial.
After having determined the label vector of image, the corresponding label of element that just can selective value is larger from the label vector is as the final label of this image.
Yet above-mentioned implementation is not optimum.Because above-mentioned implementation is from an image, the new label vector of this image only meets the match relation of this image and its k arest neighbors image when being configured.Along with the label vector of other image is updated successively, the label vector of the k of this image arest neighbors image also can change.Therefore, the new label vector of this image no longer satisfies the match relation on the characteristics of image.
The below illustrates a kind of improvement implementation of the recurrence revised law of step S3.
In the above-mentioned realization, the length of label vector is M, and each element in the vector is corresponding to a label.The label vector of k arest neighbors image by the Given Graph picture affect label vector of Given Graph picture, between each label in the label vector independently of one another.
This improved procedure has been considered between the label correlativity semantically, so that label interacts according to the correlativity between the label.
For example: " automobile " and " car " two labels have the similar meaning, and therefore, correlativity is more intense; Two labels of " automobile " and Tiger belong to respectively different classifications, and therefore, correlativity should be more weak.Wish that " automobile " label can forward impact " car " label.
Correlation calculations method between the label can adopt any suitable existing method.For example, can adopt the method for symbiosis similarity to calculate correlativity between the label.
For given image collection, suppose all independently the set that consists of of label be T={t
1..., t
M, wherein, M is the number of mutually different label.
Any two label t
iAnd t
jCorrelativity calculate as follows:
At first, calculate label t
iAnd t
jOccurrence frequency in image collection.Suppose that the picture number in the image collection is N, have label t in these images
iAnd t
jNumber be respectively p and q, so, label t
iAnd t
jFrequency be:
f(t
i)＝p/N，f(t
j)＝q/N
Then, calculate label t
iAnd t
jThe frequency that occurs simultaneously.That is: in all N of image collection image, has simultaneously label t
iAnd t
jThe shared ratio of image, be designated as f (t
i, t
j).
Then, calculate label t
iAnd t
jThe symbiosis distance:
Wherein, maximal value is got in max () expression, and minimum value is got in min () expression, and log represents to take the logarithm.
At last, calculate label t
iAnd t
jCorrelativity:
s(t
i，t
j)＝exp(-d(t
i，t
j))
Wherein, exp () is exponential function, for example the exponential function take natural logarithm e the end of as.
In the improving one's methods of the recurrence revised law of considering correlativity, step is consistent with the method for before description, and the formula that just will construct new label vector is revised as:
Wherein, S is the correlativity s (t of above-mentioned calculating
i, t
j) matrix that consists of, i, j=1 ..., M.
The below illustrates the another kind of specific implementation of step S3-batch revised law.
In the recurrence revised law, once revise the label of an image, and realize the label correction of all images by the mode of loop iteration.Revised law is realized the label correction in the image set by separating the process of optimizing in batches.
Comparatively speaking, the computation complexity of recurrence revised law is low, and computing velocity is fast.But in batches the revised law Global Optimality is good, and computation complexity is higher, and its key is to set the constraint condition that the label of construct image will satisfy.Constraint condition can be that the label configurations total error of whole image collection is minimum.In addition, constraint condition can with label between correlativity relevant.Consider the importance of original label, constraint condition also can be relevant with original label of image.
Below for example constraint condition is described.
If the initial labels of each image in image collection vector is Y
i, i=1 ..., N, N are the number of image in the image collection, and the length of each label vector is M, and M is the sum of the label that differs from one another, and the label that obtains after correction vector is f
i, i=1 ..., N, to i image, the fitting coefficient of its k arest neighbors image vector is W
Ij, j=1 ..., k.
The label configurations total error of whole image collection is:
The label of considering neotectonics should not differ too much with original label, so can add the next item up || and f
i-Y
i||
2So the label configurations total error of whole image collection is:
Can adopt the quadratic-form distance apart from dist, consider the correlativity between the label, can obtain:
Wherein, S is the label correlation matrix that correlativity obtains between the above-mentioned calculating label, and α is scale parameter, gets the number between 0 to 1, is generally obtained by experience.
Therefore, revise in batches and can obtain by separating following optimization problem:
Above-mentioned optimization problem is analyzed, is simplified, obtain following result:
Wherein, I is unit matrix, the matrix that label and initial labels consisted of after F, Y were respectively and revise.
This problem can obtain by separating the sylvester equation, and concrete solution can be by existing techniques in realizing.
It should be noted that the suitable distance that also can select other apart from dist.If distance can be led, then this optimization problem all can solve by the Gradient Descent method.Can not lead such as distance, can adopt other suitable method to find the solution this optimization problem.
After solving above-mentioned optimization problem, each image obtains a label vector, and the corresponding label of the element that selective value is larger from the label vector is as the final label of this image.
Below, with reference to the image labeling device of Fig. 2 description according to the embodiment of the invention.
Fig. 2 shows the block diagram according to the image labeling device of the embodiment of the invention.As shown in Figure 2, image labeling device 200 according to the present invention comprises: neighbour's image collection module 201, be used for each image for given image collection, in described image collection, be chosen in a plurality of other images close with described image on the characteristics of image; Fitting coefficient acquisition module 202 for the characteristics of image that passes through with the described image of characteristics of image match of described a plurality of other images, obtains a plurality of fitting coefficients of described image; And image tag constructing module 203, be used for the described a plurality of fitting coefficients according to described image, utilize the label of described a plurality of other images to construct the label of described image.
Since the processing in image labeling device according to the present invention 200 included neighbour's image collection module 201, fitting coefficient acquisition module 202 and image tag constructing module 203 respectively with the step S1-S3 of above-described image labeling method in processing similar, therefore for the sake of brevity, omit the detailed description of these modules at this.
In addition, still needing here is pointed out that, all modules, unit can be configured by the mode of software, firmware, hardware or its combination in the said apparatus.Dispose spendable concrete means or mode and be well known to those skilled in the art, do not repeat them here.In situation about realizing by software or firmware, from storage medium or network the program that consists of this software is installed to the computing machine with specialized hardware structure (for example multi-purpose computer 300 shown in Figure 3), this computing machine can be carried out various functions etc. when various program is installed.
Fig. 3 illustrates the schematic block diagram that can be used for implementing according to the computing machine of the method and apparatus of the embodiment of the invention.
In Fig. 3, CPU (central processing unit) (CPU) 301 carries out various processing according to the program of storage in the ROM (read-only memory) (ROM) 302 or from the program that storage area 308 is loaded into random access memory (RAM) 303.In RAM 303, also store as required data required when CPU 301 carries out various processing etc.CPU 301, ROM 302 and RAM 303 are connected to each other via bus 304.Input/output interface 305 also is connected to bus 304.
Following parts are connected to input/output interface 305: importation 306 (comprising keyboard, mouse etc.), output 307 (comprise display, such as cathode-ray tube (CRT) (CRT), liquid crystal display (LCD) etc., with loudspeaker etc.), storage area 308 (comprising hard disk etc.), communications portion 309 (comprising that network interface unit is such as LAN card, modulator-demodular unit etc.).Communications portion 309 is processed such as the Internet executive communication via network.As required, driver 310 also can be connected to input/output interface 305.Detachable media 311 can be installed on the driver 310 as required such as disk, CD, magneto-optic disk, semiconductor memory etc., so that the computer program of therefrom reading is installed in the storage area 308 as required.
Realizing by software in the situation of above-mentioned series of processes, such as detachable media 311 program that consists of software is being installed such as the Internet or storage medium from network.
It will be understood by those of skill in the art that this storage medium is not limited to shown in Figure 3 wherein has program stored therein, distributes separately to provide the detachable media 311 of program to the user with equipment.The example of detachable media 311 comprises disk (comprising floppy disk (registered trademark)), CD (comprising compact disc read-only memory (CD-ROM) and digital universal disc (DVD)), magneto-optic disk (comprising mini-disk (MD) (registered trademark)) and semiconductor memory.Perhaps, storage medium can be hard disk that comprises in ROM 302, the storage area 308 etc., computer program stored wherein, and be distributed to the user with the equipment that comprises them.
The present invention also proposes a kind of program product that stores the instruction code that machine readable gets.When described instruction code is read and carried out by machine, can carry out above-mentioned method according to the embodiment of the invention.
Correspondingly, being used for carrying the above-mentioned storage medium that stores the program product of the instruction code that machine readable gets is also included within of the present invention open.Described storage medium includes but not limited to floppy disk, CD, magneto-optic disk, storage card, memory stick etc.
In the above in the description to the specific embodiment of the invention, can in one or more other embodiment, use in identical or similar mode for the feature that a kind of embodiment is described and/or illustrated, combined with the feature in other embodiment, or the feature in alternative other embodiment.
Should emphasize that term " comprises/comprise " existence that refers to feature, key element, step or assembly when this paper uses, but not get rid of the existence of one or more further feature, key element, step or assembly or additional.
In addition, the time sequencing of describing during method of the present invention is not limited to is to specifications carried out, also can according to other time sequencing ground, carry out concurrently or independently.The execution sequence of the method for therefore, describing in this instructions is not construed as limiting technical scope of the present invention.
Although the above discloses the present invention by the description to specific embodiments of the invention,, should be appreciated that all above-mentioned embodiment and example all are illustrative, and not restrictive.Those skilled in the art can design various modifications of the present invention, improvement or equivalent in the spirit and scope of claims.These modifications, improvement or equivalent also should be believed to comprise in protection scope of the present invention.
Remarks
1. image labeling method comprises:
For each image in the given image collection, in described image collection, be chosen in a plurality of other images close with described image on the characteristics of image;
By the characteristics of image with the described image of characteristics of image match of described a plurality of other images, obtain a plurality of fitting coefficients of described image; And
According to described a plurality of fitting coefficients of described image, utilize the label of described a plurality of other images to construct the label of described image.
2. such as remarks 1 described image labeling method, wherein, the a plurality of fitting coefficients that obtain described image further comprise: by so that be used on the characteristics of image with Given Graph and come the error of match Given Graph picture minimum as close a plurality of other images, obtain described a plurality of fitting coefficients of described image.
3. such as remarks 1 described image labeling method, wherein, construct the label of described image to satisfy predetermined constraint condition.
4. such as remarks 3 described image labeling methods, wherein, described constraint condition comprises that the label configurations total error of whole image collection is minimum.
5. such as remarks 3 described image labeling methods, wherein, described constraint condition is relevant with the correlativity between the label.
6. such as remarks 3 described image labeling methods, wherein, described constraint condition is relevant with original label of described image.
7. such as remarks 1 described image labeling method, the label of wherein constructing described image further comprises:
At random or an image in the described image collection of select progressively;
Utilization comes the label of the selected image of match corresponding to the label of a plurality of other images of selected image with the fitting coefficient of correspondence; And
Repeat above-mentioned steps, until be in the described image collection each image configuration label.
8. such as remarks 7 described image labeling methods, wherein the label of the selected image of match further comprises: based on the correlativity between the label, utilize the label corresponding to a plurality of other images of selected image, come the label of the selected image of match with the fitting coefficient of correspondence.
9. such as the described image labeling method of one of remarks 1-8, the wherein said linear fit that fits to, described label is expressed with vector form.
10. image labeling device comprises:
Neighbour's image collection module is used for each image for given image collection, is chosen in a plurality of other images close with described image on the characteristics of image in described image collection;
The fitting coefficient acquisition module for the characteristics of image that passes through with the described image of characteristics of image match of described a plurality of other images, obtains a plurality of fitting coefficients of described image; And
The image tag constructing module is used for the described a plurality of fitting coefficients according to described image, utilizes the label of described a plurality of other images to construct the label of described image.
11. such as remarks 10 described image labeling devices, wherein, described fitting coefficient acquisition module obtains described a plurality of fitting coefficients of described image by so that be used on the characteristics of image with Given Graph and come the error of match Given Graph picture minimum as close a plurality of other images.
12. such as remarks 10 described image labeling devices, wherein, described image tag constructing module is constructed the label of described image to satisfy predetermined constraint condition.
13. such as remarks 12 described image labeling devices, wherein, described constraint condition comprises that the label configurations total error of whole image collection is minimum.
14. such as remarks 12 described image labeling devices, wherein, described constraint condition is relevant with the correlativity between the label.
15. such as remarks 12 described image labeling devices, wherein, described constraint condition is relevant with original label of described image.
16. such as the described image labeling device of one of remarks 10-15, wherein, the described linear fit that fits to, described label is expressed with vector form.
Claims (10)
1. image labeling method comprises:
For each image in the given image collection, in described image collection, be chosen in a plurality of other images close with described image on the characteristics of image;
By the characteristics of image with the described image of characteristics of image match of described a plurality of other images, obtain a plurality of fitting coefficients of described image; And
According to described a plurality of fitting coefficients of described image, utilize the label of described a plurality of other images to construct the label of described image.
2. image labeling method as claimed in claim 1, wherein, the a plurality of fitting coefficients that obtain described image further comprise: by so that be used on the characteristics of image with Given Graph and come the error of match Given Graph picture minimum as close a plurality of other images, obtain described a plurality of fitting coefficients of described image.
3. image labeling method as claimed in claim 1 wherein, is constructed the label of described image to satisfy predetermined constraint condition.
4. image labeling method as claimed in claim 3, wherein, described constraint condition comprises that the label configurations total error of whole image collection is minimum.
5. image labeling method as claimed in claim 3, wherein, described constraint condition is relevant with the correlativity between the label.
6. image labeling method as claimed in claim 3, wherein, described constraint condition is relevant with original label of described image.
7. image labeling method as claimed in claim 1, the label of wherein constructing described image further comprises:
At random or an image in the described image collection of select progressively;
Utilization comes the label of the selected image of match corresponding to the label of a plurality of other images of selected image with the fitting coefficient of correspondence; And
Repeat above-mentioned steps, until be in the described image collection each image configuration label.
8. image labeling device comprises:
Neighbour's image collection module is used for each image for given image collection, is chosen in a plurality of other images close with described image on the characteristics of image in described image collection;
The fitting coefficient acquisition module for the characteristics of image that passes through with the described image of characteristics of image match of described a plurality of other images, obtains a plurality of fitting coefficients of described image; And
The image tag constructing module is used for the described a plurality of fitting coefficients according to described image, utilizes the label of described a plurality of other images to construct the label of described image.
9. image labeling device as claimed in claim 8, wherein, described image tag constructing module is constructed the label of described image to satisfy predetermined constraint condition.
10. image labeling device as claimed in claim 9, wherein, described constraint condition comprises that the label configurations total error of whole image collection is minimum; Perhaps described constraint condition is relevant with the correlativity between the label; Perhaps described constraint condition is relevant with original label of described image.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210084554.5A CN103365850B (en) | 2012-03-27 | 2012-03-27 | Image labeling method and image labeling device |
| JP2013064274A JP6102401B2 (en) | 2012-03-27 | 2013-03-26 | Image labeling method and apparatus |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210084554.5A CN103365850B (en) | 2012-03-27 | 2012-03-27 | Image labeling method and image labeling device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103365850A true CN103365850A (en) | 2013-10-23 |
| CN103365850B CN103365850B (en) | 2017-07-14 |
Family
ID=49367225
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210084554.5A Expired - Fee Related CN103365850B (en) | 2012-03-27 | 2012-03-27 | Image labeling method and image labeling device |
Country Status (2)
| Country | Link |
|---|---|
| JP (1) | JP6102401B2 (en) |
| CN (1) | CN103365850B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090290802A1 (en) * | 2008-05-22 | 2009-11-26 | Microsoft Corporation | Concurrent multiple-instance learning for image categorization |
| CN101963995A (en) * | 2010-10-25 | 2011-02-02 | 哈尔滨工程大学 | Image marking method based on characteristic scene |
| CN102142089A (en) * | 2011-01-07 | 2011-08-03 | 哈尔滨工程大学 | Semantic binary tree-based image annotation method |
| CN102236798A (en) * | 2011-08-01 | 2011-11-09 | 清华大学 | Image matching method and device |
| CN102298606A (en) * | 2011-06-01 | 2011-12-28 | 清华大学 | Random walking image automatic annotation method and device based on label graph model |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101388638B1 (en) * | 2008-06-06 | 2014-04-24 | 구글 인코포레이티드 | Annotating images |
| US8775424B2 (en) * | 2010-01-26 | 2014-07-08 | Xerox Corporation | System for creative image navigation and exploration |
-
2012
- 2012-03-27 CN CN201210084554.5A patent/CN103365850B/en not_active Expired - Fee Related
-
2013
- 2013-03-26 JP JP2013064274A patent/JP6102401B2/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090290802A1 (en) * | 2008-05-22 | 2009-11-26 | Microsoft Corporation | Concurrent multiple-instance learning for image categorization |
| CN101963995A (en) * | 2010-10-25 | 2011-02-02 | 哈尔滨工程大学 | Image marking method based on characteristic scene |
| CN102142089A (en) * | 2011-01-07 | 2011-08-03 | 哈尔滨工程大学 | Semantic binary tree-based image annotation method |
| CN102298606A (en) * | 2011-06-01 | 2011-12-28 | 清华大学 | Random walking image automatic annotation method and device based on label graph model |
| CN102236798A (en) * | 2011-08-01 | 2011-11-09 | 清华大学 | Image matching method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| JP2013206470A (en) | 2013-10-07 |
| CN103365850B (en) | 2017-07-14 |
| JP6102401B2 (en) | 2017-03-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110162593B (en) | Search result processing and similarity model training method and device | |
| US11636147B2 (en) | Training neural networks to perform tag-based font recognition utilizing font classification | |
| US11727243B2 (en) | Knowledge-graph-embedding-based question answering | |
| CN111966917B (en) | Event detection and summarization method based on pre-training language model | |
| CN111444320B (en) | Text retrieval method and device, computer equipment and storage medium | |
| Demir et al. | A novel active learning method in relevance feedback for content-based remote sensing image retrieval | |
| US11741157B2 (en) | Propagating multi-term contextual tags to digital content | |
| Wu et al. | Positive and unlabeled multi-graph learning | |
| US8825744B2 (en) | Active image tagging | |
| CN108460011B (en) | Entity concept labeling method and system | |
| US9208218B2 (en) | Methods and apparatuses for generating search expressions from content, for applying search expressions to content collections, and/or for analyzing corresponding search results | |
| CN112819023B (en) | Sample set acquisition method, device, computer equipment and storage medium | |
| CN111539197B (en) | Text matching method and device, computer system and readable storage medium | |
| CN105393264A (en) | Interactive segment extraction in computer-human interactive learning | |
| CN106951498A (en) | Text clustering method | |
| US20130204835A1 (en) | Method of extracting named entity | |
| US20230401426A1 (en) | Prediction method, prediction apparatus and program | |
| Jamaati et al. | Text mining by Tsallis entropy | |
| CN115455171A (en) | Method, device, equipment and medium for mutual retrieval and model training of text videos | |
| Wang et al. | Knowledge mining with scene text for fine-grained recognition | |
| CN111709475B (en) | N-gram-based multi-label classification method and device | |
| Qian et al. | Boosted multi-modal supervised latent Dirichlet allocation for social event classification | |
| CN112417147A (en) | Method and device for selecting training samples | |
| CN112364198A (en) | Cross-modal Hash retrieval method, terminal device and storage medium | |
| CN103377381A (en) | Method and device for identifying content attribute of image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| 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: 20170714 Termination date: 20180327 |