CN107633048A - A kind of image labeling discrimination method and system - Google Patents
A kind of image labeling discrimination method and system Download PDFInfo
- Publication number
- CN107633048A CN107633048A CN201710834040.XA CN201710834040A CN107633048A CN 107633048 A CN107633048 A CN 107633048A CN 201710834040 A CN201710834040 A CN 201710834040A CN 107633048 A CN107633048 A CN 107633048A
- Authority
- CN
- China
- Prior art keywords
- marked
- image set
- replicated
- collection
- image
- 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
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This application discloses a kind of image labeling discrimination method and system, this method to include:According to image set to be marked, acquisition is replicated image set and copy pattern image set, described to be replicated the subset that image set is the image set to be marked, is replicated image set described in the copy pattern image set duplication and obtains;It is replicated corresponding to image set to have marked described in obtaining respectively and is replicated that image set is corresponding with the duplicating image collection to have marked copy pattern image set;Marked described in judging and be replicated whether image set meets the first preparatory condition with the similitude for having marked copy pattern image set, if it is, process decision chart has cheating as annotation process.The present invention has been marked by judgement to be replicated image set and has marked whether the similitude of copy pattern image set meets the first preparatory condition to judge to whether there is cheating in annotation process, set cheating criterion, when image data amount is very big, remains to efficiently differentiate and whether there is cheating in annotation process.
Description
Technical field
The present invention relates to image labeling field, more particularly to a kind of image labeling discrimination method and system.
Background technology
In machine learning field, image is manually labeled in model training with playing vital work in terms of model evaluation
With.On the one hand, during model training, the image manually marked is usually as the data of model training, image labeling quality
The performance of model can be directly affected;On the other hand, during model evaluation, the image manually marked is usually as algorithm
The normative reference that can be assessed, image labeling quality also play decisive role.
When the view data for needing to mark is privately owned, the mass-rent of the databases such as similar ImageNet and COCO should not be used
Pattern is labeled, it will usually privately owned view data is contracted out into professional team and is labeled, now, then needs to differentiate image mark
Whether there is cheating to evaluate mark quality during note, especially when image data amount is very big, more need efficiently to differentiate
Annotation process whether there is cheating.
The content of the invention
In view of this, it is an object of the invention to provide a kind of image labeling discrimination method and system, can efficiently differentiate
It whether there is cheating during image labeling.Its concrete scheme is as follows:
A kind of image labeling discrimination method, including:
According to image set to be marked, acquisition is replicated image set and copy pattern image set, and the image set that is replicated is described
The subset of image set to be marked, the copy pattern image set replicate described in be replicated image set and obtain;
It is replicated described in obtaining respectively corresponding to image set and has marked that to be replicated image set corresponding with the duplicating image collection
The copy pattern image set of mark;
Marked described in judging and be replicated whether image set meets first with the similitude for having marked copy pattern image set
Preparatory condition, if it is, process decision chart has cheating as annotation process.
Preferably, marked described in the judgement and be replicated image set and the similitude for having marked copy pattern image set and be
The no process for meeting the first preparatory condition, including:
Described marked is obtained respectively to be replicated image set and marked that duplicating image collection is corresponding actually to mark collection with described
Close, obtain being replicated the actual mark set of image set and the actual mark of copy pattern image set is gathered;
It is corresponding with the actual mark set of the copy pattern image set that the actual mark set of image set is replicated described in obtaining respectively
Class label collection and tab area collection;
According to class label collection and tab area collection, respectively to it is described marked be replicated image set with it is described marked it is multiple
The image set that charts carries out similarity measurement, obtains corresponding overlapping tab area area ratio and overlapping tab area number ratio;
Image set and the overlapping tab area area ratio for having marked copy pattern image set are replicated using described marked,
Corresponding area parameters are calculated, and has been marked described in utilization and has been replicated image set and the weight for having marked copy pattern image set
Folded tab area number ratio, calculates corresponding number parameter;
Obtaining includes the first predetermined threshold value of preset area threshold value and predetermined number threshold value, and has marked and answered described in judgement
Drawing image set causes the area parameters not by the default face with the duplicating image concentration that marked with the presence or absence of image
Within the areal extent that product threshold value determines and/or the number parameter is not in the quantitative range determined by the predetermined number threshold value
Within, it is replicated image set and the similar sexual satisfaction institute for having marked copy pattern image set if it is, having been marked described in judging
State the first preparatory condition.
Preferably, further comprise:
Image set is obscured in acquisition, wherein, the common factor for obscuring image set and the image set to be marked is empty set;
Obscure whether image set meets the second preparatory condition described in judgement, if it is, process decision chart exists as annotation process
Cheating.
Preferably, the process whether image set meets the second preparatory condition is obscured described in the judgement, including:
The theoretical mark set that image set is obscured described in acquisition is gathered with actual mark, obtain obscuring theoretical mark set with
Obscure actual mark set;
Theoretical mark set and the MD5 code collections for obscuring actual mark set are obscured described in obtaining respectively;
Obscure described in judgement the MD5 code collections of theoretical mark set and obscure that actual mark gathers the MD5 code collections and
Whether collection element number is with the difference for obscuring the theoretical element number for marking the MD5 code collections gathered more than the second default threshold
Value, if it is, obscuring image set described in judging meets second preparatory condition.
Preferably, further comprise:
Acquisition has marked total collection, it is described marked total collection for it is described marked image set, described marked is replicated image
Collection, the union for obscuring image set for having marked copy pattern image set and having marked;
Mark whether total collection meets the 3rd preparatory condition described in judging, if it is, process decision chart exists as annotation process
Cheating.
Preferably, the process whether total collection meets the 3rd preparatory condition has been marked described in the judgement, including:
The actual mark set always collected has been marked described in obtaining, has obtained the actual mark set of total collection;
Obtain MD5 code collections corresponding to the actual mark set of total collection;
Judge that the element number of the actual mark set of total collection and total collection are actual and mark the corresponding MD5 code collections of set
Element number difference whether more than the 3rd predetermined threshold value, if it is, judge it is described marked always to collect meet the described 3rd
Preparatory condition.
Preferably, further comprise:
Obtain total collection to be marked, total collection to be marked is the image set to be marked, the image set, described of being replicated
Copy pattern image set and the union for obscuring image set;
Judge that total collection to be marked has marked whether total collection meets the 4th preparatory condition with described, if it is, judging
Cheating be present in image labeling process.
Preferably, it is described to judge total collection to be marked and the mistake for having marked total collection and whether having met the 4th preparatory condition
Journey, including:
Obtain the element number to be marked to be marked always collected;
The MD5 code collections to be marked always collected are obtained respectively and have marked the MD5 code collections always collected with described, are obtained to be marked
Always collection always collects MD5 codes with having marked MD5 codes;
The MD5 codes to be marked always collection and the common factor element number for having marked MD5 codes and always having collected are obtained, obtains MD5 codes
Common factor element number;
Judge the element number to be marked with the difference of the MD5 codes common factor element number whether more than the 4th default threshold
It is worth, if it is, judging that total collection to be marked meets the 4th preparatory condition with the always collection that marked.
Correspondingly, the present invention also provides a kind of image labeling identification system, including:
First acquisition module, for being replicated image set and copy pattern image set, the quilt according to image set to be marked, acquisition
Copy pattern image set is the subset of the image set to be marked, and the copy pattern image set obtains to be replicated image set described in duplication
's;
Second acquisition module, for obtain respectively it is described be replicated corresponding to image set to have marked be replicated image set and institute
State and copy pattern image set has been marked corresponding to duplicating image collection;
First judge module, for judging that described marked is replicated image set and the phase for having marked copy pattern image set
Whether meet the first preparatory condition like property, if it is, process decision chart has cheating as annotation process.
Preferably, further comprise:
3rd acquisition module, obscure image set for obtaining, wherein, it is described to obscure image set and the image set to be marked
Common factor be empty set;
Second judge module, for judge it is described obscure whether image set meets the second preparatory condition, if it is, judging
Cheating be present in image labeling process.
Image labeling discrimination method disclosed by the invention and system, image set and copy pattern image set are replicated by obtaining,
Judgement, which has marked, to be replicated image set and has marked whether the similitude of copy pattern image set meets the first preparatory condition to judge to mark
It whether there is cheating during note.Because copy pattern image set is to replicate to obtain by being replicated image set, so the two
Theoretical annotation results should be highly similar, and works as the similitude for the actual annotation results for being replicated image set and copy pattern image set not
When enough high, that is, when meeting the first preparatory condition, judge cheating be present in annotation process.Sentence because the present invention sets cheating
Calibration is accurate, so when image data amount is very big, remains to efficiently differentiate and whether there is cheating in annotation process.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the required accompanying drawing used in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
The embodiment of invention, for those of ordinary skill in the art, on the premise of not paying creative work, can also basis
The accompanying drawing of offer obtains other accompanying drawings.
Fig. 1 is a kind of flow chart of image labeling discrimination method disclosed in the embodiment of the present invention;
Fig. 2 is that the judgement disclosed in the embodiment of the present invention has marked the phase for being replicated image set and having marked copy pattern image set
Whether meet the flow chart of the first preparatory condition like property;
Fig. 3 is that the flow chart whether image set meets the second preparatory condition is obscured in the judgement disclosed in the embodiment of the present invention;
Fig. 4 is that the judgement disclosed in the embodiment of the present invention has marked the flow chart whether total collection meets the 3rd preparatory condition;
Fig. 5 is to judge total collection to be marked disclosed in the embodiment of the present invention and marked whether total collection meets the 4th default article
The flow chart of part;
Fig. 6 is a kind of structure chart of image labeling identification system disclosed in the embodiment of the present invention;
Fig. 7 is the structural representation of another image labeling identification system disclosed in the embodiment of the present invention.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation describes, it is clear that described embodiment is only part of the embodiment of the present invention, rather than whole embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained every other under the premise of creative work is not made
Embodiment, belong to the scope of protection of the invention.
Shown in Figure 1 the embodiment of the invention discloses a kind of image labeling discrimination method, Fig. 1 is the embodiment of the present invention
A kind of flow chart of disclosed image labeling discrimination method, comprises the following steps:
Step S11:According to image set to be marked, acquisition is replicated image set and copy pattern image set, described to be replicated image
Collect for the subset of the image set to be marked, the copy pattern image set replicate described in be replicated image set and obtain;
In the embodiment of the present invention, Ω is usedthImage set to be marked is represented,Wherein N1To be just whole
Number, N1Represent ΩthIn element number, i.e., picture number to be marked;Use ΩrExpression is replicated image set,ΩrFor ΩthSubset, ΩrIn image can be from ΩthMiddle arbitrary extracting, meet
, use ΩreCopy pattern image set is represented,Due to ΩreTo replicate ΩrObtain, so ΩrWith
ΩreElement number be all N2, wherein N2For positive integer, and N2≤N1。
Step S12:It is replicated corresponding to image set to have marked described in obtaining respectively and is replicated image set and the copy pattern
Copy pattern image set has been marked corresponding to image set;
Specifically, useExpression is replicated image set ΩrCorresponding marked is replicated image set,
Wherein M2For positive integer, representIn element number;WithRepresent copy pattern image set ΩreIt is corresponding to have marked duplicating image
Collection,Wherein M3For positive integer, representIn element number.It should be noted that for upper
State M2With M3, there is M2≤M1And M3≤M1, wherein, M1Represent image set Ω to be markedthIt is corresponding to have marked image setElement
Number.
Step S13:Judge whether described marked is replicated image set and the similitude for having marked copy pattern image set
Meet the first preparatory condition, if it is, process decision chart has cheating as annotation process.
Wherein, the first preparatory condition is related to similitude, and can according to be actually needed set area threshold and
The criterion of amount threshold.
Image labeling discrimination method disclosed in the embodiment of the present invention, image set and copy pattern image set are replicated by obtaining,
Judgement, which has marked, to be replicated image set and has marked whether the similitude of copy pattern image set meets the first preparatory condition to judge to mark
It whether there is cheating during note.Because copy pattern image set is to replicate to obtain by being replicated image set, so the two
Theoretical annotation results should be highly similar, and works as the similitude for the actual annotation results for being replicated image set and copy pattern image set not
When enough high, that is, when meeting the first preparatory condition, judge cheating be present in annotation process.Sentence because the present invention sets cheating
Calibration is accurate, so when image data amount is very big, remains to efficiently differentiate and whether there is cheating in annotation process.
It is replicated image set to judging to have marked in above-mentioned steps S13 below and is with the similitude for having marked copy pattern image set
The no process for meeting the first preparatory condition, which is made, to be illustrated, shown in Figure 2, and Fig. 2 is sentencing disclosed in the embodiment of the present invention
Disconnected marked is replicated the flow chart whether image set meets the first preparatory condition with having marked the similitude of copy pattern image set, bag
Include following steps:
Step S21:Described marked is obtained respectively to be replicated image set and marked the corresponding reality of duplicating image collection with described
Border mark set, obtain being replicated the actual mark set of image set and the actual mark of copy pattern image set is gathered;
Specifically, setting marked set in i-th image with this marked set it is corresponding actually mark gather in
I-th of result one-to-one relationship be present.Image set is replicated to have markedGather with being replicated the actual mark of image setExemplified by, if representing filename with name, haveThe attribute beyond filename can certainly be used
Show corresponding relation, this embodiment of the present invention is not limited, and this setting may be equally applicable for other collection and close, hereinafter
It will not be described in further detail.
In the way of above-mentioned setting, acquisition, which has marked, is replicated image setIt is corresponding to be replicated the actual mark of image set
Collection is combined intoCopy pattern image set is markedThe corresponding actual mark collection of copy pattern image set is combined into
It should be noted that acquisition disclosed above is replicated the actual mark set of image setIt is actual with copy pattern image set
Mark setAcquisition modes suitable for situation of the image set to be marked from copy pattern image set different files, i.e. quilt
The filename of copy pattern image set and copy pattern image set can with it is identical when.
If image set to be marked is with copy pattern image set when same file presss from both sides, i.e. is replicated image set and duplicating image
When the filename of collection can not repeat, the image that duplicating image is concentrated needs renaming.Now, due to being replicated image set with replicating
The filename of two content identical pictures is different corresponding in image set, so to obtain copy pattern image set, then needs to pass through
The MD5 codes of image set are replicated, compares in this document folder and hunts out the copy pattern image set for possessing identical MD5 codes.
Step S22:The actual mark set of image set and the actual mark of the copy pattern image set are replicated described in obtaining respectively
Class label collection corresponding to set and tab area collection;
Specifically, setting, which has marked the marked content of set, includes class label collection and tab area collection, and classification mark
Corresponding relation be present with tab area collection in label collection.WithExemplified by,Annotation results content beWherein
Class label collection isTab area collection istiRepresentIn class
Distinguishing label number, classificationCorresponding tab area collection isWherein, ni
Represent classificationComprising tab area number.It is rightSetting similarly, will not be repeated here.
Step S23:According to class label collection and tab area collection, respectively to it is described marked be replicated image set with it is described
Mark copy pattern image set and carried out similarity measurement, obtain corresponding overlapping tab area area ratio and overlapping tab area number
Than;
Specifically, similarity measurement is:Image set is replicated for having markedWith having marked copy pattern image setIf
Consider certain mark imageWithCorresponding annotation resultsWithIn identical category(i=1,2 ..., M2, j=1,
2,…,ti, k=1,2 ..., ni), it is defined as follows:
RepresentIn i-th mark image j-th of classification corresponding to k-th of tab area area, pi,jRepresentIn i-th mark image j-th of classification corresponding to tab area number;
RepresentIn i-th mark image j-th of classification corresponding to k-th of tab area area, qi,jRepresentIn i-th mark image j-th of classification corresponding to tab area number;
RepresentWithIn i-th mark image j-th of classification corresponding to k-th of tab area faying surface
Product, si,jRepresentWithIn i-th mark image j-th of classification corresponding to overlapping tab area number.
According to the above description,Overlapping tab area area ratioAnd overlapping tab area number ratio Overlapping tab area area ratioAnd overlapping tab area number ratio
Step S24:Image set and the overlapping marked area for having marked copy pattern image set are replicated using described marked
Domain area ratio, calculates corresponding area parameters, and utilizes described marked to be replicated image set and described marked duplication
The overlapping tab area number ratio of image set, calculates corresponding number parameter;
Wherein, area parametersNumber parameter
Step S25:Obtaining includes the first predetermined threshold value of preset area threshold value and predetermined number threshold value, and described in judgement
Mark is replicated image set and causes the area parameters not by institute with the presence or absence of image with the duplicating image concentration that marked
State within the areal extent of preset area threshold value determination and/or the number parameter is not being determined by the predetermined number threshold value
Within quantitative range, the similar of copy pattern image set has been marked to described if it is, having been marked described in judging and being replicated image set
First preparatory condition described in sexual satisfaction.
Specifically, judgeWithIn whether there is imageWithMakeIt is unsatisfactory forAnd/or
MakeIt is unsatisfactory for(i=1,2 ..., M2, j=1,2 ..., ti, k=1,2 ..., ni, andIf it is present judgeWithThe preparatory condition of similar sexual satisfaction first.Wherein,With
When closer 1, illustrate that similitude is higher, and preset area threshold valueWith predetermined number threshold valueValue can basis
It is actually needed and is set,WithValue closer to 1, then illustrate that judging standard is stricter.
In order to strengthen identification result, image labeling discrimination method, further comprises disclosed in the embodiment of the present invention:
Image set is obscured in acquisition, wherein, the common factor for obscuring image set and the image set to be marked is empty set;
Specifically, i.e. obscure the element for not including image set to be marked in image set.
Obscure whether image set meets the second preparatory condition described in judgement, if it is, process decision chart exists as annotation process
Cheating.
Specifically, shown in Figure 3, Fig. 3 is that the judgement disclosed in the embodiment of the present invention obscures whether image set meets the
The flow chart of two preparatory conditions, comprises the following steps:
Step S31:The theoretical mark set that image set is obscured described in acquisition is gathered with actual mark, obtains obscuring theoretical mark
Note set is gathered with obscuring actual mark;
Specifically, Ω is usedfaExpression does not include ΩthElement obscures image set,Wherein, N3
For positive integer, and N3≤N1, N3Represent ΩfaIn element number, N1Represent ΩthIn element number, obscure theoretical mark set
ForThen mark and obscured image set and beIt is corresponding to obscure reality
Mark collection is combined intoWherein, M4≤M1And M4For positive integer, M4RepresentIn element number.
Step S32:Theoretical mark set and the MD5 code collections for obscuring actual mark set are obscured described in obtaining respectively;
In the embodiment of the present invention, set set X MD5 codes asSet E element number is card (E), and this sets
It is fixed to be applied to all embodiments hereinafter simultaneously.
Then obscure theoretical mark set DfaMD5 code collections beObscure actual mark setMD5 code collections be
Step S33:The MD5 code collections of theoretical mark set and the MD5 for obscuring actual mark set are obscured described in judgement
Whether the union element number of code collection is with the difference for obscuring the theoretical element number for marking the MD5 code collections gathered more than second
Predetermined threshold value, if it is, obscuring image set described in judging meets second preparatory condition.
Specifically, the judgment formula of the second preparatory condition is(Γ2For positive integer), if satisfied, then judging that obscuring image set meets the second preparatory condition.Need to illustrate
, due to obscuring image set ΩfaNot comprising image set Ω to be markedthImage, so in theory, obscuring image set Ωfa's
Image should not all mark, i.e.,AndIllustrate pair
Obscure image set ΩfaActual annotation results.When occurring cheating in annotation process, i.e., to obscuring image set ΩfaCarry out
During mark,Result be positive integer, wherein Γ2, can for threshold parameter
It is actually needed and is set with basis, Γ2Closer to 0, illustrate that judging standard is stricter.
Again in order to strengthen identification result, image labeling discrimination method, further comprises disclosed in the embodiment of the present invention:
Acquisition has marked total collection, it is described marked total collection for it is described marked image set, described marked is replicated image
Collection, the union for obscuring image set for having marked copy pattern image set and having marked;
Mark whether total collection meets the 3rd preparatory condition described in judging, if it is, process decision chart exists as annotation process
Cheating.
Specifically, shown in Figure 4, Fig. 4 is that the judgement disclosed in the embodiment of the present invention has marked whether total collection meets the
The flow chart of three preparatory conditions, comprises the following steps:
Step S41:The actual mark set always collected has been marked described in obtaining, has obtained the actual mark set of total collection;
Wherein, useExpression has marked total collection,Then total collection
Actual mark collection is combined intoWherein M >=0 and M are integer, and M is representedIn element number.
Step S42:Obtain MD5 code collections corresponding to the actual mark set of total collection;
Specifically, MD5 code collections corresponding to the actual mark set of total collection are
Step S43:Judge that the element number of the actual mark set of total collection is corresponding with total actual mark set of collection
MD5 code collections element number difference whether more than the 3rd predetermined threshold value, if it is, judging described to have marked total collection satisfaction
3rd preparatory condition.
In the embodiment of the present invention, the judgment formula of the 3rd preparatory condition is(Γ3For
Positive integer), if satisfied, then judging that having marked total collection meets the 3rd preparatory condition.Wherein, Γ3It can be carried out according to being actually needed
Setting, Γ3Smaller, then explanation differentiates definitely stricter.It should be noted that the embodiment of the present invention is actually to overall mark
Mark behavior during note is differentiated, if cheating in annotation process be present, for example, different images be present carries out phase
During with situation about marking,It will be less than
Again in order to strengthen identification result, image labeling discrimination method, further comprises disclosed in the embodiment of the present invention:
Obtain total collection to be marked, total collection to be marked is the image set to be marked, the image set, described of being replicated
Copy pattern image set and the union for obscuring image set;
Judge that total collection to be marked has marked whether total collection meets the 4th preparatory condition with described, if it is, judging
Cheating be present in image labeling process.
Specifically, shown in Figure 5, Fig. 5 is that the total collection to be marked of the judgement disclosed in the embodiment of the present invention is total with having marked
Whether collection meets the flow chart of the 4th preparatory condition, comprises the following steps:
Step S51:Obtain the element number to be marked to be marked always collected;
Specifically, it is to be marked always to integrate as Ω=Ωth∪Ωre∪Ωfa, corresponding element number to be marked is card (Ω).
Step S52:The MD5 code collections to be marked always collected are obtained respectively and have marked the MD5 code collections always collected with described, are obtained
Always collection always collects MD5 codes to be marked with having marked MD5 codes;
Wherein, always collection is MD5 codes to be markedHaving marked MD5 codes, always collection is
Step S53:The MD5 codes to be marked always collection and the common factor element number for having marked MD5 codes and always having collected are obtained, is obtained
To MD5 code common factor element numbers;
Specifically, MD5 codes common factor element number is
Step S54:Judge the difference of the element number to be marked and the MD5 codes common factor element number whether more than the
Four predetermined threshold values, if it is, judging that total collection to be marked meets the 4th preparatory condition with the always collection that marked.
In the embodiment of the present invention, the judgment formula of the 4th preparatory condition is
(Γ4For positive integer), if satisfied, then judging that total collection to be marked meets the 4th preparatory condition with having marked total collection.Need what is illustrated
It is that the embodiment of the present invention is also in annotation process, Ω amount of images is verified, when Ω is changed, such as when Ω's
When amount of images is reduced,Amount of images also accordingly reduce,AsWith the number of Ω identical images
Amount, thenTo mark the front and rear quantity by modification image, wherein Γ4Can be according to reality
Need to be set, Γ4Closer to 0, illustrate that judging standard is stricter.
It should be noted that in actual discrimination process, can be according to precision needs be differentiated, from disclosed in previous embodiment
Required mode is selected to differentiate in four kinds of specific embodiments of image labeling discrimination method.For example, when to annotation results requirement
When not high, can only judge second, third, the 4th preparatory condition, when to annotation results require it is very high when, it is proposed that four default bars
Part all judges, wherein the deterministic process of the first preparatory condition is the most complicated, but judges precision highest.
Correspondingly, the embodiment of the present invention also provides a kind of image labeling identification system, shown in Figure 6, and Fig. 6 is the present invention
A kind of structure chart of image labeling identification system, the system include disclosed in embodiment:
First acquisition module 61, for according to image set to be marked, acquisition to be replicated image set and copy pattern image set, described
The subset that image set is the image set to be marked is replicated, the copy pattern image set obtains to be replicated image set described in duplication
's;
Second acquisition module 62, for obtain respectively it is described be replicated corresponding to image set marked be replicated image set with
Copy pattern image set has been marked corresponding to the duplicating image collection;
First judge module 63, for judging that described marked is replicated image set and the copy pattern image set that marked
Whether similitude meets the first preparatory condition, if it is, process decision chart has cheating as annotation process.
Further, shown in Figure 7, Fig. 7 is another image labeling identification system disclosed in the embodiment of the present invention
Structural representation, the system include:
3rd acquisition module 71, obscure image set for obtaining, wherein, it is described to obscure image set and the image to be marked
The common factor integrated is empty set;
Second judge module 72, for judge it is described obscure whether image set meets the second preparatory condition, if it is, sentencing
Determine image labeling process and cheating be present.
Foregoing implementation is may be referred on the more specifical course of work of modules in above-mentioned image labeling identification system
Corresponding contents disclosed in example, are no longer repeated herein.
Image labeling discrimination method and system disclosed in the embodiment of the present invention, image set and copy pattern are replicated by obtaining
Image set, judgement, which has marked, to be replicated image set and has marked whether the similitude of copy pattern image set meets the first preparatory condition to sentence
It whether there is cheating in disconnected annotation process.Because copy pattern image set is to replicate to obtain by being replicated image set, so two
The theoretical annotation results of person should be highly similar, and similar to the actual annotation results of copy pattern image set when being replicated image set
Property it is not high enough when, that is, when meeting the first preparatory condition, judge annotation process in cheating be present.Because the present invention sets work
Disadvantage criterion, so when image data amount is very big, remains to efficiently differentiate and whether there is cheating in annotation process.
Finally, it is to be noted that, herein, such as first and second or the like relational terms be used merely to by
One entity or operation make a distinction with another entity or operation, and not necessarily require or imply these entities or operation
Between any this actual relation or order be present.Moreover, term " comprising ", "comprising" or its any other variant meaning
Covering including for nonexcludability, so that process, method, article or equipment including a series of elements not only include that
A little key elements, but also the other element including being not expressly set out, or also include for this process, method, article or
The intrinsic key element of equipment.In the absence of more restrictions, the key element limited by sentence "including a ...", is not arranged
Except other identical element in the process including the key element, method, article or equipment being also present.
Image labeling discrimination method provided by the present invention and system are described in detail above, it is used herein
Specific case is set forth to the principle and embodiment of the present invention, and the explanation of above example is only intended to help and understands this
The method and its core concept of invention;Meanwhile for those of ordinary skill in the art, according to the thought of the present invention, specific
There will be changes in embodiment and application, in summary, this specification content should not be construed as to the present invention's
Limitation.
Claims (10)
- A kind of 1. image labeling discrimination method, it is characterised in that including:According to image set to be marked, acquisition is replicated image set and copy pattern image set, and the image set that is replicated is waited to mark to be described The subset of image set is noted, being replicated image set described in the copy pattern image set duplication obtains;It is replicated corresponding to image set to have marked described in obtaining respectively and is replicated image set and the duplicating image collection is corresponding Mark copy pattern image set;Marked described in judging and be replicated whether image set meets that first is default with the similitude for having marked copy pattern image set Condition, if it is, process decision chart has cheating as annotation process.
- 2. according to the method for claim 1, it is characterised in that marked described in the judgement be replicated image set with it is described Whether the similitude for having marked copy pattern image set meets the process of the first preparatory condition, including:Described marked is obtained respectively to be replicated image set and marked that duplicating image collection is corresponding actually to mark set with described, is obtained Gather to the actual mark set of image set is replicated with the actual mark of copy pattern image set;The actual mark set of image set and the actual mark corresponding class of set of the copy pattern image set are replicated described in obtaining respectively Distinguishing label collection and tab area collection;According to class label collection and tab area collection, described marked is replicated by image set copy pattern is marked with described respectively Image set carries out similarity measurement, obtains corresponding overlapping tab area area ratio and overlapping tab area number ratio;Image set and the overlapping tab area area ratio for having marked copy pattern image set are replicated using described marked, is calculated Go out corresponding area parameters, and marked described in utilization and be replicated image set and the overlapping mark for having marked copy pattern image set Areal ratio is noted, calculates corresponding number parameter;Obtaining includes the first predetermined threshold value of preset area threshold value and predetermined number threshold value, and has been marked described in judgement and be replicated figure Image set causes the area parameters not by the preset area threshold with the duplicating image concentration that marked with the presence or absence of image Be worth determine areal extent within and/or the number parameter not the quantitative range determined by the predetermined number threshold value it It is interior, marked if it is, having been marked described in judging and being replicated image set to described described in the similar sexual satisfaction of copy pattern image set First preparatory condition.
- 3. method according to claim 1 or 2, it is characterised in that further comprise:Image set is obscured in acquisition, wherein, the common factor for obscuring image set and the image set to be marked is empty set;Obscure whether image set meets the second preparatory condition described in judgement, if it is, process decision chart is practised fraud as annotation process is present Behavior.
- 4. according to the method for claim 3, it is characterised in that obscure whether image set meets that second is pre- described in the judgement If the process of condition, including:The theoretical mark set that image set is obscured described in acquisition is gathered with actual mark, obtains obscuring theoretical mark set with obscuring Actual mark set;Theoretical mark set and the MD5 code collections for obscuring actual mark set are obscured described in obtaining respectively;The MD5 code collections of theoretical mark set and the MD5 code collections and element of set for obscuring actual mark set are obscured described in judgement Whether the difference of prime number mesh and the element number of the MD5 code collections for obscuring theoretical mark set is more than the second predetermined threshold value, such as Fruit is that then obscuring image set described in judgement meets second preparatory condition.
- 5. according to the method for claim 3, it is characterised in that further comprise:Acquisition has marked total collection, it is described marked total collection be described in marked image set, described marked is replicated image set, institute State the union for obscuring image set for having marked copy pattern image set and having marked;Mark whether total collection meets the 3rd preparatory condition described in judging, if it is, process decision chart is practised fraud as annotation process is present Behavior.
- 6. according to the method for claim 5, it is characterised in that marked whether total collection meets that the 3rd is pre- described in the judgement If the process of condition, including:The actual mark set always collected has been marked described in obtaining, has obtained the actual mark set of total collection;Obtain MD5 code collections corresponding to the actual mark set of total collection;Judge the actual member for marking the corresponding MD5 code collections of set of the element number of the actual mark set of total collection and total collection Whether prime number purpose difference is more than the 3rd predetermined threshold value, if it is, total collection has been marked described in judging meets that the described 3rd is default Condition.
- 7. according to the method for claim 5, it is characterised in that further comprise:Obtain total collection to be marked, total collection to be marked is the image set to be marked, described is replicated image set, the duplication Image set and the union for obscuring image set;Judge that total collection to be marked has marked whether total collection meets the 4th preparatory condition with described, if it is, process decision chart picture Cheating be present in annotation process.
- 8. the method according to claim 11, it is characterised in that described to judge that total collection to be marked has marked always with described Whether collection meets the process of the 4th preparatory condition, including:Obtain the element number to be marked to be marked always collected;The MD5 code collections to be marked always collected are obtained respectively and have marked the MD5 code collections always collected with described, obtain MD5 codes to be marked Total collection always collects with having marked MD5 codes;The MD5 codes to be marked always collection and the common factor element number for having marked MD5 codes and always having collected are obtained, obtains MD5 codes common factor Element number;Whether the difference of the element number to be marked and the MD5 codes common factor element number is judged more than the 4th predetermined threshold value, If it is, judge that total collection to be marked meets the 4th preparatory condition with the always collection that marked.
- A kind of 9. image labeling identification system, it is characterised in that including:First acquisition module, for according to image set to be marked, acquisition to be replicated image set and copy pattern image set, described to be replicated Image set is the subset of the image set to be marked, and being replicated image set described in the copy pattern image set duplication obtains;Second acquisition module, for obtain respectively it is described be replicated corresponding to image set marked be replicated image set with it is described multiple Copy pattern image set has been marked corresponding to imaged collection;First judge module, for judging that described marked is replicated image set and the similitude for having marked copy pattern image set Whether first preparatory condition is met, if it is, process decision chart has cheating as annotation process.
- 10. system according to claim 9, it is characterised in that further comprise:3rd acquisition module, obscure image set for obtaining, wherein, the friendship for obscuring image set and the image set to be marked Integrate as empty set;Second judge module, for judge it is described obscure whether image set meets the second preparatory condition, if it is, process decision chart picture Cheating be present in annotation process.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710834040.XA CN107633048B (en) | 2017-09-15 | 2017-09-15 | Image annotation identification method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710834040.XA CN107633048B (en) | 2017-09-15 | 2017-09-15 | Image annotation identification method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107633048A true CN107633048A (en) | 2018-01-26 |
CN107633048B CN107633048B (en) | 2021-02-26 |
Family
ID=61101463
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710834040.XA Active CN107633048B (en) | 2017-09-15 | 2017-09-15 | Image annotation identification method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107633048B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110335251A (en) * | 2019-05-31 | 2019-10-15 | 上海联影智能医疗科技有限公司 | Quantization device, method, equipment and the storage medium of image analysis method |
CN110348507A (en) * | 2019-07-03 | 2019-10-18 | 创新奇智(南京)科技有限公司 | A kind of anti-cheating method of image labeling, system and electronic equipment |
CN111209621A (en) * | 2019-12-31 | 2020-05-29 | 深圳市华阳国际工程设计股份有限公司 | Cross-view dimension marking and copying method, terminal and storage medium |
CN111666936A (en) * | 2019-03-08 | 2020-09-15 | 北京市商汤科技开发有限公司 | Labeling method, labeling device, labeling system, electronic equipment and storage medium |
CN114820456A (en) * | 2022-03-30 | 2022-07-29 | 图湃(北京)医疗科技有限公司 | Image processing method and device |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101064009A (en) * | 2006-04-28 | 2007-10-31 | 夏普株式会社 | Image processing apparatus, image forming apparatus, image reading apparatus and image processing method |
CN102045162A (en) * | 2009-10-16 | 2011-05-04 | 电子科技大学 | Personal identification system of permittee with tri-modal biometric characteristic and control method thereof |
US20140160534A1 (en) * | 2012-12-06 | 2014-06-12 | Konica Minolta, Inc. | Image processing method, image forming apparatus and image forming system |
CN103984738A (en) * | 2014-05-22 | 2014-08-13 | 中国科学院自动化研究所 | Role labelling method based on search matching |
CN105023163A (en) * | 2015-06-23 | 2015-11-04 | 杭州沃朴物联科技有限公司 | Anti-counterfeiting system based on chaotic graphic label and method |
CN105975980A (en) * | 2016-04-27 | 2016-09-28 | 百度在线网络技术(北京)有限公司 | Method of monitoring image mark quality and apparatus thereof |
CN107633066A (en) * | 2017-09-21 | 2018-01-26 | 北京金山安全软件有限公司 | Information display method and device, electronic equipment and storage medium |
JP2018136900A (en) * | 2017-02-24 | 2018-08-30 | 東芝情報システム株式会社 | Sentence analysis device and sentence analysis program |
-
2017
- 2017-09-15 CN CN201710834040.XA patent/CN107633048B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101064009A (en) * | 2006-04-28 | 2007-10-31 | 夏普株式会社 | Image processing apparatus, image forming apparatus, image reading apparatus and image processing method |
CN102045162A (en) * | 2009-10-16 | 2011-05-04 | 电子科技大学 | Personal identification system of permittee with tri-modal biometric characteristic and control method thereof |
US20140160534A1 (en) * | 2012-12-06 | 2014-06-12 | Konica Minolta, Inc. | Image processing method, image forming apparatus and image forming system |
CN103984738A (en) * | 2014-05-22 | 2014-08-13 | 中国科学院自动化研究所 | Role labelling method based on search matching |
CN105023163A (en) * | 2015-06-23 | 2015-11-04 | 杭州沃朴物联科技有限公司 | Anti-counterfeiting system based on chaotic graphic label and method |
CN105975980A (en) * | 2016-04-27 | 2016-09-28 | 百度在线网络技术(北京)有限公司 | Method of monitoring image mark quality and apparatus thereof |
JP2018136900A (en) * | 2017-02-24 | 2018-08-30 | 東芝情報システム株式会社 | Sentence analysis device and sentence analysis program |
CN107633066A (en) * | 2017-09-21 | 2018-01-26 | 北京金山安全软件有限公司 | Information display method and device, electronic equipment and storage medium |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111666936A (en) * | 2019-03-08 | 2020-09-15 | 北京市商汤科技开发有限公司 | Labeling method, labeling device, labeling system, electronic equipment and storage medium |
CN110335251A (en) * | 2019-05-31 | 2019-10-15 | 上海联影智能医疗科技有限公司 | Quantization device, method, equipment and the storage medium of image analysis method |
CN110335251B (en) * | 2019-05-31 | 2021-09-17 | 上海联影智能医疗科技有限公司 | Quantization apparatus, method, device and storage medium for image analysis method |
CN110348507A (en) * | 2019-07-03 | 2019-10-18 | 创新奇智(南京)科技有限公司 | A kind of anti-cheating method of image labeling, system and electronic equipment |
CN111209621A (en) * | 2019-12-31 | 2020-05-29 | 深圳市华阳国际工程设计股份有限公司 | Cross-view dimension marking and copying method, terminal and storage medium |
CN114820456A (en) * | 2022-03-30 | 2022-07-29 | 图湃(北京)医疗科技有限公司 | Image processing method and device |
Also Published As
Publication number | Publication date |
---|---|
CN107633048B (en) | 2021-02-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107633048A (en) | A kind of image labeling discrimination method and system | |
CN104318562B (en) | A kind of method and apparatus for being used to determine the quality of the Internet images | |
CN105118048B (en) | The recognition methods of reproduction certificate picture and device | |
CN104143079B (en) | The method and system of face character identification | |
CN109657665A (en) | A kind of invoice batch automatic recognition system based on deep learning | |
CN103325122B (en) | Based on the pedestrian retrieval method of Bidirectional sort | |
CN106469299A (en) | A kind of vehicle search method and device | |
CN105955890B (en) | A kind of generation method and device of functional test case | |
CN108052980A (en) | Air quality grade detection method based on image | |
CN103984941A (en) | Face recognition checking-in method and device thereof | |
JP2007026386A (en) | Image search system and method | |
CN108830267A (en) | A kind of method and system goed over examination papers based on image recognition | |
CN103049446A (en) | Image retrieving method and device | |
CN103221978A (en) | Person detection device and person detection method | |
CN105095475B (en) | Imperfect attribute based on two-graded fusion marks pedestrian recognition methods and system again | |
CN107392929A (en) | A kind of intelligent target detection and dimension measurement method based on human vision model | |
CN102306415B (en) | Portable valuable file identification device | |
CN109948549A (en) | OCR data creation method, device, computer equipment and storage medium | |
CN107992783A (en) | Face image processing process and device | |
CN106485710A (en) | Method and device for detecting wrong component | |
CN110175990A (en) | Quality of human face image determination method, device and computer equipment | |
CN108897821A (en) | The method and device of automatically generated data conclusion | |
CN110377659A (en) | A kind of intelligence chart recommender system and method | |
CN104751406A (en) | Method and device used for blurring image | |
CN108334589A (en) | A kind of facial skin care product recommendation method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |