CN103064858A - Method and apparatus for objectionable image detection in social networking websites - Google Patents
Method and apparatus for objectionable image detection in social networking websites Download PDFInfo
- Publication number
- CN103064858A CN103064858A CN2011103237617A CN201110323761A CN103064858A CN 103064858 A CN103064858 A CN 103064858A CN 2011103237617 A CN2011103237617 A CN 2011103237617A CN 201110323761 A CN201110323761 A CN 201110323761A CN 103064858 A CN103064858 A CN 103064858A
- Authority
- CN
- China
- Prior art keywords
- probability
- image
- bad
- user
- photograph album
- 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 embodiment of the invention relates to a method and equipment for detecting bad images in a social networking website. Specifically, disclosed is a method for use in poor image detection of a social networking SNS site, comprising: performing analysis on a first image in the SNS website to determine a first probability that the first image is a bad image; in response to the first probability being within a predetermined range, performing analysis on at least one second image in the album to which the first image belongs to determine at least one second probability that the at least one second image is a bad image; and adjusting the first probability according to at least one second probability, wherein the adjusted first probability is used for judging whether the first image is a bad image. A corresponding apparatus is also disclosed. According to the embodiment of the invention, the accuracy and stability of poor image detection can be improved.
Description
Technical field
Embodiments of the present invention relate generally to network information technology field, more specifically, relate to the method and apparatus for the bad image detection of social networking website.
Background technology
Along with the development of Internet technology, a lot of and image-related network services has appearred.Utilize these services, the user can and be stored in the network the image issue, and can be online to various operations such as image browsing, editor, annotations and comments, comments.Especially, along with the development of social networks (SNS), the user can develop and manage one or more catalogues that are specifically designed to image in the SNS website in recent years.This image directory is commonly referred to " photograph album " in the art.At present, existing hundreds of millions of image is published and is stored in such photograph album.
A problem that is closely related with online images serve is the detection to bad image.Here alleged " bad image " refers to comprise and violates concerned countries or zonal law, rules or other may produce the image of dysgenic content.For example, some tissues or individual may and propagate by online images serve storage and comprise for example image of pornographic, violence, terror or other harmful contents.In the face of magnanimity at line image, it is unpractical that bad image is carried out hand inspection.Therefore, some technology of bad image have appearred being intended to automatically detect and filter.
In known prior art, detection scheme can be divided into two classes generally automatically.The first kind depends on color and the texture in the image.For example, can set up complexion model or texture model by means such as training in advance.For any image to be detected, can utilize computer vision and image processing techniques to determine area of skin color in the image, then infer that based on the previous complexion model of determining human body exposes degree, whether be pornographic image thereby detect this image.The Equations of The Second Kind scheme depends on the attitude of personage in the image.Can set up the prior model of bad attitude, then utilize it to determine whether comprise same or analogous posture in the given image to be detected, thereby determine whether this image contains the contents such as pornographic, violence.This two classes scheme all exists a lot of distortion and evolution separately, also has the trial that two class schemes are combined.In addition, also can be used as the detection that supplementary means promotes bad image with image-related text message (for example, title, description, comment etc.).
Yet there is obvious limitation in existing scheme.At first, computer vision and image processing techniques can't guarantee accurately, detect reliably the content that comprises in the image at present.For example, in depending on the scheme of complexion model, the various factorss such as illumination, exposure, shooting angle may produce uncertain impact to testing result.Equally, because people's attitude is ever-changing, put into practice requirement based on the matching process of priori attitude mode still can't satisfying aspect the precision.Secondly, in the prior art, ignored image publisher's information and speciality take image as the detection on basis merely, thereby may cause than undetected and flase drop more frequently.The problems referred to above have larger instability so that existing bad image detects automatically.
Therefore, need in this area a kind of more accurate, stable and effectively detect the scheme of bad image.
Summary of the invention
In view of above problem, the present invention proposes a kind of method and apparatus of the bad image detection that is used for social networking website of novelty.
On the one hand, the invention provides a kind of method of in the bad image detection of social networks SNS website, using.The method comprises: to the first image execution analysis in the described SNS website, to determine that described the first image is as the first probability of bad image; Be in the preset range in response to described the first probability, at least one the second image execution analysis in the photograph album under described the first image, to determine that described at least one second image is as at least one second probability of bad image; And regulate described the first probability according to described at least one second probability, described the first probability after wherein regulating will be used to judge whether described the first image is bad image.
Aspect another, the invention provides a kind of equipment for detect bad image in social networks SNS website.This equipment comprises: the first probability is determined device, and configuration is used for the first image execution analysis to described SNS website, to determine that described the first image is as the first probability of bad image; The second probability is determined device, configuration is used for being in the preset range in response to described the first probability, to at least one the second image execution analysis in the photograph album under described the first image, to determine that described at least one second image is as at least one second probability of bad image; And the first probability regulating device, configuration is used for regulating described the first probability according to described at least one second probability, and described the first probability after wherein regulating will be used to judge whether described the first image is bad image.
Embodiments of the present invention take full advantage of the characteristics of SNS website.At first, the image in the SNS website is organized take photograph album as unit usually, and the statistical study of SNS website is shown: the user is issue and the same or analogous image of storage class in same photograph album often.In addition, user behavior analysis also shows: often have similar behavior pattern and custom between the good friend of SNS website.Thus, in the bad image detection of SNS website, can also include the SNS incidence relation between the user in consideration.Be attached in the bad image detection by the These characteristics with the SNS website, can effectively improve the Stability and veracity of detection.
Description of drawings
By reading with reference to the accompanying drawings detailed description hereinafter, above-mentioned and other purposes of embodiment of the present invention, the feature and advantage easy to understand that will become.In the accompanying drawings, show some embodiments of the present invention in exemplary and nonrestrictive mode, wherein:
Fig. 1 shows the process flow diagram of the method 100 of using according to one exemplary embodiment in the bad image detection of SNS website;
Fig. 2 shows the block diagram of the equipment 200 that uses according to one exemplary embodiment in the bad image detection of SNS website; And
Fig. 3 shows the block diagram of the computer system 300 that is fit to put into practice embodiment of the present invention.
In each accompanying drawing, identical or corresponding label represents identical or corresponding part.
Embodiment
Below with reference to some illustrative embodiments principle of the present invention and spirit are described.Should be appreciated that providing these embodiments only is for those skilled in the art can being understood better and then realize the present invention, and be not to limit the scope of the invention by any way.
As indicated above, according to the embodiment of the present invention, when detecting bad image, take full advantage of image organizational feature and/or user behavior pattern feature in the SNS website.The useful information that provides by these features combines with bad image detection, can effectively improve the Stability and veracity of detection.
In order more clearly to explain principle of the present invention and spirit, below the definition of some terms of relating to herein of given first.At first, as used herein term " social networks " or " social network sites " or " social networking system " be point to interested in the special object or just together the people of " saunter " the web website of virtual community is provided.After registration and login, the member of social networks can communicate by voice, chat, instant message, video conference and blog etc.Social networks provides the method that contacts other members to the member usually.Social networks can also be as the medium of in person meeting.
In addition, term " social network members " or abbreviation " member " are to point to the user that social networks has carried out registration and may also pass through relevant authentication as used herein.Notice that in the description relevant with social networks, " member " and " user " is used interchangeably.
In addition, as used herein term " good friend " refer to social networks the member by social networks a plurality of users that are formed with each other connection, related or relation.(but this is not necessarily) that connection in the social networks is normally two-way, so term " good friend " may depend on reference system.Connection between the user can be direct connection; Yet some embodiment of social networks allows the indirect joint via one-level or multistage connection.In addition, term " good friend " is not to necessarily require the user to be actually friend in actual life, and it only represents the relation in the social networks.
Term " good friend's progression " refers to the number of the connection between the good friend in the social networks as used herein.For example, if set up between the user A of social networks and other user B direct connection is arranged, then user A and B are the one-level good friends; Be connected if but another user C sets up to have to be connected to have with user A foundation with other user B, then user A and C are the secondary good friends; By that analogy.One-level good friend also can be called " directly good friend ", and other good friend of good friend's level also can be called " indirectly good friend ".
After the term definition on provided, below with reference to the accompanying drawings, explain in detail principle of the present invention and spirit in conjunction with some embodiments.
At first with reference to figure 1, show the process flow diagram of the method 100 of in the bad image detection of social networks SNS website, using according to one exemplary embodiment.
As described in Figure, after method 100 beginnings, at step S102, a Given Graph in the SNS website is looked like to analyze, to determine that this image is as the probability of bad image.Clear and easy for what describe, the image that step S102 place is analyzed is called " the first image ", and is that the probability of bad image is called " the first probability " with the first image.
According to the embodiment of the present invention, the first image can be any image in the SNS website.For example, the first image can be that the user has just uploaded to the SNS website image to be released is arranged, or the image of selecting at random from the existing image of SNS website.And for example, the keeper of SNS website can specify doubtful user with bad behavior, and this first image can be the image that user under a cloud has, issues, checks or comments on.Notice that these only are several examples, are not intended to limit scope of the present invention.
According to the embodiment of the present invention, can any suitable method/algorithm to the analysis of the first image.For example, for example can comprise the analysis of the first image and to analyze the content that this image comprises, that is, and content-based analysis.As example, content-based analysis can comprise the analysis to color of image.Take the pornographic image detection as example, can utilize the complexion model of precondition extract with matching image in the area of skin color that comprises, to determine whether comprise a large amount of exposed skins in the image.In addition, content-based analysis can also comprise the attitude of the object (people or other objects) that comprises in the image is analyzed.This for example can be by realizing to the extraction of geometric properties and with the coupling of prior model.Be appreciated that, no matter be that color model or attitude mode all can utilize the whole bag of tricks to realize, various statistical models such as Markov model (HMM), pivot analysis (PCA), support vector machine (SVM), perhaps neuroid, etc.
Alternatively or additionally, can comprise the text of analyzing with the first image correlation connection to the analysis of the first image.This class text can include but not limited to following one or more: the title of image, and the parameter (for example, the parameter that EXIF information is entrained) that image is self-contained, about the comment of image, the user is to the description of image, etc.Various text analyzing methods all can be combined with the present invention, for example sentence segmentation, keyword extraction and coupling, grammatical analysis, semantic analysis, etc.
Notice that content and/or text analyzing at step S102 place to the first image can utilize any means to realize, no matter be at present known or in the future exploitation of this area.What above enumerate only is several examples, and scope of the present invention is not restricted in this regard.
According to the analysis to the first image, can determine the probability that the first image is bad image.Particularly, according to the embodiment of the present invention, can utilize various suitable technological means will be quantified as for the analysis result of the first image bad image probability, that is, and the first probability.
Take the analysis of color-based as example, a kind of simple strategy is: with the colour of skin in image shared area as the first probability.And for example, use a model or the embodiment of neuroid in, can quantitatively determine the first probability according to the color of the first image and/or the matching degree of object attitude and corresponding prior model.
Equally, for the text based analysis, can according to the keyword that extracts, grammer and/or matching degree semantic and predetermined keyword, grammer and/or semanteme, quantitatively determine the first probability.In addition, can be to bad keyword, grammer and/or the semantic rank of dividing, for example, this keyword of the higher expression of rank, grammer and/or semanteme more might represent harmful content, and correspondingly the first probable value is also just larger.
Especially, the first image is being carried out simultaneously in the situation of content analysis and text analyzing, can utilized various suitable means that the determined probability of content analysis and the determined probability of text analyzing are carried out combination.For example, can to various suitable computings such as the numerical applications arithmetic mean of this two classes probability, weighted mean, geometric mean, summations, determine that thus the first image is the first probability of bad image.
Should be appreciated that above cited probability determines that method is exemplary, only be intended to provide the enlightenment of determining the first probability by graphical analysis.According to the concrete image analysis method that adopts, can determine that quantitatively the first image is the first probability of bad image with any technological means alternative and/or that add.Scope of the present invention is unrestricted in this regard.
Next, method 100 proceeds to step S104, and whether determined the first probability is within the preset range at this determining step S102 place.
According to the embodiment of the present invention, this preset range can be automatic or manual definite, and can be configurable.For example, the keeper of SNS website can be rule of thumb or statistics lower threshold and the upper limit threshold of bad image probability are set.At this moment, the probit range between lower threshold and the upper limit threshold can be used as " preset range ".
If determine not to be in (branch's "No") in the preset range with the first probability of the first image correlation connection at step S104 place, then method finishes.Be appreciated that, in this case, the first image might (for example directly be judged to be legal image, in the situation of the first probability less than lower threshold) or bad image is (for example, in the situation of the first probability greater than upper limit threshold), and can carry out corresponding subsequent operation.These aspects are not 100 problems of concerns of method, do not consist of limiting the scope of the invention yet.
On the other hand, if determine that at step S104 place the first probability is in (branch's "Yes") in this preset range, then can think: directly be that legal image or bad image may be not accurate enough or reliable with the first spectral discrimination.At this moment, the first image can be considered to " doubtful bad image ".
Correspondingly, method 100 proceeds to step S106, this to the first image under at least one other image execution analysis in the photograph album, to determine that this or these image is as at least one probability of bad image.For the purpose of clear, the handled image of step S106 be called " the second image ", and be that the probability of bad image is called " the second probability " with the second image.
As indicated above, in the SNS website, image is in most of the cases organized take photograph album as unit.And the photograph album information under each image is known or obtainable in the SNS website.Thus, at step S106, can select one or more images as the second image the photograph album under the first image.According to the embodiment of the present invention, this selection can be carried out according to various standards, includes but not limited to: select the second image of the predetermined number in this photograph album, select at random, select all images in this photograph album, etc.These only are examples, and scope of the present invention is unrestricted in this regard.
According to the embodiment of the present invention, at step S106 place, can utilize any method of developing known or future at present at least one the second image execution analysis.For example, can carry out content-based analysis and/or for the analysis of related text to the second image.Then can be at least one second probability of bad image according at least one second image of this Analysis deterrmination.Should be appreciated that the analysis of the second image and be similar to analysis and the probability to the first image that refer step S102 above describes according to the process of this Analysis deterrmination the second probability and determine.Correlative detail does not repeat them here.
After this, method 100 proceeds to step S108, according at least one second probability that step S106 determines, is adjusted in the first probability that step S102 place determines at this.
As indicated above, the statistical study of SNS website is shown: the same or analogous image of containing type often in the same photograph album in the SNS website.This is ubiquitous feature in the SNS website, has universality and stability, is a kind of inherent law of SNS website.Thus, be that the probability (that is, the second probability) of bad image is regulated the first probability by utilizing other images in the identical photograph album, can judge more exactly whether bad image of the first image.
Generally speaking, the overall principle of utilizing the second probability to regulate the first probability is: if other images in the same photograph album (namely, at least one second image) is that the probability of bad image is generally higher, then can correspondingly improves the first probability that the first image is bad image; Otherwise, generally lower if other images in the same photograph album (that is, at least one second image) are the probability of bad image, then can correspondingly reduce the first probability that the first image is bad image.
Particularly, at step S108 place, such as can be average to the first probability and at least one the second probability applied arithmetic, the various algorithms that are averaging such as geometric mean, weighted mean.The first probability after the result of gained can be used as regulating.Be appreciated that when the second probability is generally higher the first probability after regulating like this may will be higher than its initial value, vice versa.
As another example more accurately, the probability of establishing the first image and be bad image is P
1, and the number that is located at the second image of analyzing among the step S106 is n (n is natural number), the second image is that the probability of bad image is respectively P
21, P
22..., P
2nAt step S108 place, can be according to function
P
1＝f(P
21，P
22，...，P
2n)
Regulate the first probability P
1, P wherein
1The dependent variable of function f, P
21, P
22..., P
2nBe the independent variable of function f, and function f is increasing function (that is, P
1With P
21, P
22..., P
2nIncrease and increase.Angle from mathematics, in interval or its sub-range, may be used to embodiments of the present invention for the various function f of increasing function at independent variable, include but not limited to: direct proportion function, linear increasing function, index be greater than 1 logarithmic function, specific power function, specific exponential function, etc.And, also can first to carrying out specific computing, for example average.Then, with the independent variable of gained unique value as f.
In regulating the first probability, except with the value of at least one the second probability itself as the independent variable, can also utilize the distributed intelligence of its numerical value.For example, a kind of simple feasible program is: if P
21, P
22..., P
2nIn greater than the predetermined number that outnumbers of given threshold value, then with the first probability P
1Increase progressively certain specific value.Otherwise, if P
21, P
22..., P
2nIn less than the predetermined number that outnumbers of given threshold value, then with the first probability P
1Certain specific value of successively decreasing.Certainly can be to P
21, P
22..., P
2nThe numeric distribution situation carry out more accurate statistical calculations and measurement, and regulate accordingly P
1
Notice that above-described all only is example, any other suitable concrete technological means can be used to all realize that the first probability based on the second probability regulates.Scope of the present invention is unrestricted in this regard.
By the Adjustment operation of step S108, for the first probability that originally is in the preset range, its numerical value can be increased or decreased according at least one the second probability.The first probability after regulating like this can be used to then judge that whether the first image is bad image (for example, can by comparing to realize with predetermined threshold).In this way, the information of the first image itself has not only been considered in the judgement of the first image, but also the information that will belong to other images in the same photograph album is included consideration in, can improve the precision and stability of judgement.
Alternatively, according to some embodiment of the present invention, be in the situation of preset range at the first probability with the first image correlation connection, can also be further with reference to the user who has the first image (for example, the user who issues or store this first image) information is in order to further improve the Stability and veracity of bad image detection.
In such embodiment, method 100 can proceed to step S110, determines to have the user of the first image at this and is bad user's probability.For the purpose of clear, this probability is called as " the 3rd probability ".
According to the embodiment of the present invention, user probability (that is, the 3rd probability) of belonging to bad user can have any suitable initial value.For example, this initial value can default setting be zero or any other numerical value, by the keeper of SNS website arrange or according to suitable algorithm by the machine Lookup protocol.
According to some embodiment of the present invention, in order to determine the currency of the 3rd probability, can check whether this user is in the bad user list.Here said " bad user list " also can be described as " blacklist ", usually administered and maintained by the SNS operator, be used for being recorded in the SNS website and had the user's of bad behavior (usually more than once) information (for example, user's ID or user name, etc.).User's bad behavior for example can comprise following one or more: once issued or propagated bad image, once checked and/or commented on the bad image of other people issue, or improper or illegal behavior is thought in other any SNS websites.Be found to be in the bad user list if having the user of the first image, then the 3rd probability can be made as higher value (even be made as 100%, that is, directly the user is regarded as bad user).
In addition, the definite of the 3rd probability can also be based on the record of the historical behavior of this user in the SNS website.Particularly, in the SNS website, usually not can the user be engaged in once or the situation of several times bad behavior of minority under just the user is directly added bad user list.Therefore, a kind of situation that may exist is: although the user carried out bad behavior in the past, not yet be added in bad user list or the blacklist.For tackling this situation, according to the embodiment of the present invention, determining of the 3rd probability can be with the historical behavior record of user in the SNS website as one of foundation.Be appreciated that the bad behavior number of times in user's historical record is more, the numerical value of the 3rd probability is also correspondingly larger.
Alternatively or additionally, the 3rd probability determines whether to be in the bad user list based on this user at least one good friend in the SNS website.User behavior analysis to the SNS website shows: there are similarity at least to a certain extent in behavior pattern, custom and tendency with SNS website user of good friend's relation.Therefore, the number that user is in the good friend in the bad user list is more, and the probability (that is, the 3rd probability) that this user itself belongs to bad user is also larger.
The mode of above-described several definite the 3rd probability all is exemplary, is not intended to limit scope of the present invention.Any other suitable mode all can be combined with embodiments of the present invention.And, no matter adopt which kind of mode, all can adopt any suitable technological means quantitatively to calculate the value of the 3rd probability.For example, (for example can design any suitable increasing function, those exemplary increasing functions mentioned above), the number of the user's bad behavior in the historical record and/or the number that the user is in the good friend in the blacklist are determined as dependent variable with the 3rd probability as independent variable.Above enumerated the example of some feasible increasing functions, do not repeated them here.
Especially, according to some embodiment, the probability that user is bad user can upgrade in time.In other words, for any one given user, whenever its probability that belongs to bad user changes, just cover previous value with the value after changing, thereby realize dynamically updating and following the trail of of the 3rd probability.
Next, method 100 proceeds to step S112, further regulates the first probability at this 3rd probability that utilizes step S110 place to determine.Utilize the 3rd probability that the adjusting of the first probability is similar to step S108 place and utilize the second probability to the adjusting of the first probability, do not repeat them here.
Can be used to judge whether bad image (for example, by with the comparison of predetermined threshold) of the first image through the first probability after further regulating.In this way, in bad image detection, not only consider the information of other images in the same photograph album, but also also included user's information in consideration.This can further improve the Stability and veracity of bad image detection.
Alternatively, in some embodiments, method 100 can proceed to step S114, the photograph album this manages the first image according to the first probability of determining and the second probability under.Be appreciated that " the first probability " that step S114 place refers to passed through the adjusting of step S108 and/or step S112.
According to the embodiment of the present invention, can determine that the photograph album under the first image is the probability (for the purpose of clear, being referred to as " the 4th probability ") of bad photograph album.For example, according to some embodiment, can will be averaging algorithm application in the first probability and the second probability, and with acquired results as the 4th probability.Alternatively, also can add up in the first probability and the second probability number that surpasses predetermined threshold, and the 4th probability is set to the increasing function (that is, the doubtful bad image in the photograph album is more, and then this photograph album is that the probability of bad photograph album is higher) of this number.These only are examples, are not intended to limit scope of the present invention.
After determining the 4th probability, can judge accordingly that whether the photograph album under the first image is bad photograph album (for example, by with relatively the realizing of predetermined threshold)." bad photograph album " described here refers to specially or mainly be used to store the photograph album of bad image.Paying special attention to, is not necessarily to be carried out by method 100 to this operation of judgement of photograph album.In other words, can oneself judge according to the 4th probability whether photograph album is bad photograph album by method 100, also can utilize method 100 the 4th definite probability to carry out this type of judgement by other method or process.
If photograph album is judged as bad photograph album according to determined the 4th probability, according to some embodiment, all images that this photograph album can be comprised directly are judged to be bad image.In other words, though in this photograph album not by analysis image also will be judged as bad image.In this way, can utilize the feature that exists of image in the SNS website, save significantly time and the resources costs that is associated with the individual images analysis.
In addition, method 100 can proceed to step S116 alternatively, manages the user who has the first image according to the result of determination of the first image.Pay special attention to, whether be bad image although the first probability can be used to judge the first image, yet judge that this operation itself is not necessarily to be carried out by method 100.In other words, can oneself judge according to the first probability whether the first image is bad image by method 100, also can utilize method 100 the first definite probability to carry out this type of judgement by other method or process.
Return step S116, in some embodiments, the counter that is associated with each user can be set, be used for this user of record in the bad behavior of SNS website.At this moment, if the first image is confirmed as bad image according to the first probability, then can increase progressively the counter that is associated with this user.The value that is appreciated that this counter can be used to judge whether photograph album is bad photograph album.For example, according to some embodiment, if this counter surpassed predetermined threshold number, then this user can be judged to be bad user.In this case, according to some illustrative embodiments of the present invention, this user for example can be added in bad user list or the blacklist.
By reference to the accompanying drawings to the description of method 100, please pay special attention to the following aspects for above.At first, each step of record can be carried out and/or executed in parallel according to different orders in the method 100.For example, be appreciated that between step (S106, S108) and the step (S110, S112), the dependence on the out-of-order also between step (S110, S112) and the step S114.It can be carried out or executed in parallel according to different orders.And method 100 can also comprise additional step and/or omit the step shown in carrying out.
Secondly, the fundamental purpose of method 100 is to determine the first probability and determine alternatively second, third and/or the 4th probability.These probability can be used to corresponding judgement, but decision itself is not necessarily to be carried out by method 100.In other words, these decision are dispensable for solving institute of the present invention problems of concern.
In addition, be appreciated that based on method 100 determined the first probability and be that a kind of automatic machine is judged to the judgement that the first image is done.The result of this judgement can directly be used as net result, also can be submitted to the keeper of SNS website for example to carry out manual confirmation.After final certain image of affirmation belongs to bad image, can take various treatment measures equally, for example directly delete this image, warning user, limited subscriber authority, closed user account, etc.Protection scope of the present invention is all unrestricted in these areas.
Below with reference to Fig. 2, it shows the block diagram according to the equipment 200 that uses of exemplary embodiment of the invention in the bad image detection of social networks SNS website.Equipment 200 for example can reside at the SNS website and be responsible in the server of image detection and filtration, or its other modes are associated with this server.
As shown in the figure, according to the embodiment of the present invention, equipment 200 comprises: the first probability is determined device, and configuration is used for the first image execution analysis to described SNS website, to determine that described the first image is as the first probability of bad image; The second probability is determined device, configuration is used for being in the preset range in response to described the first probability, to at least one the second image execution analysis in the photograph album under described the first image, to determine that described at least one second image is as at least one second probability of bad image; And the first probability regulating device, configuration is used for regulating described the first probability according to described at least one second probability, and described the first probability after wherein regulating will be used to judge whether described the first image is bad image.
According to some embodiment of the present invention, equipment 200 further comprises: the 3rd probability is determined device, configuration is used for being in the described preset range in response to described the first probability, determine to have the user of described the first image and be bad user's the 3rd probability, wherein said the first probability regulating device further configuration is used for regulating described the first probability according to described the 3rd probability.
According to some embodiment of the present invention, described the 3rd probability determine device configuration be used for based on following at least one determine described the 3rd probability: whether described user is in bad user list; Whether described user at least one good friend in described SNS website is in the described bad user list; And the record of the historical behavior of described user in described SNS website.
According to some embodiment of the present invention, equipment 200 further comprises: the user management device, configuration is used for being judged as bad image in response to described the first image according to described the first probability after regulating and increases progressively the counter that is associated with the user who has described the first image, described counter is used for recording described user in the bad behavior of described SNS website, and the value of wherein said counter will be used to judge whether described user is bad user.
According to some embodiment of the present invention, equipment 200 further comprises: the 4th probability is determined device, and configuration is used for according to described the first probability and described at least one second probability, determines that described photograph album is the 4th probability of bad photograph album.
According to some embodiment of the present invention, equipment 200 further comprises: photograph album management devices, configuration are used for being judged as bad photograph album in response to described photograph album according to described the 4th probability, and all spectral discriminations that described photograph album is comprised are bad image.
According to some embodiment of the present invention, to a Given Graph look like to carry out described analysis comprise following at least one: analyze the content that described Given Graph picture comprises; And analysis and the described Given Graph text that looks like to be associated.The device that is used for analysis image can be positioned at equipment 200 inside, that is, be the device of equipment 200.Alternatively, the device for analysis image also can be positioned at outside the equipment 200 also independent with it.
Note, for the purpose of clear, the sub-device that optional device and each device comprise is not shown in Fig. 2.Yet, each device that should be appreciated that in the equipment 200 record respectively with the method 100 of describing with reference to figure 1 in each step.Thus, above for the operation of method 100 descriptions among Fig. 1 and the device that feature is equally applicable to equipment 200 and wherein comprises, do not repeat them here.
It is also understood that equipment 200 can utilize variety of way to realize.For example, in some embodiments, equipment 200 can utilize software and/or firmware to realize.Alternatively or additionally, equipment 200 can partially or fully be realized based on hardware.For example, equipment 200 can be implemented as integrated circuit (IC) chip or special IC (ASIC).Equipment 200 also can be implemented as SOC (system on a chip) (SOC).Other modes known or in the future exploitation also are feasible now, and scope of the present invention is unrestricted in this regard.
Fig. 3 shows the schematic block diagram of the computer system that is suitable for putting into practice embodiment of the present invention.For example, computer system shown in Figure 3 can be used as the server of being responsible for bad image detection and filtration in the SNS website, and comprises above-described equipment 200 or associated.
As shown in Figure 3, computer system can comprise: CPU (CPU (central processing unit)) 301, RAM (random access memory) 302, ROM (ROM (read-only memory)) 303, system bus 304, hard disk controller 305, keyboard controller 306, serial interface controller 307, parallel interface controller 308, display controller 309, hard disk 310, keyboard 311, serial external unit 312, parallel external unit 313 and display 314.In these equipment, with system bus 304 coupling CPU 301, RAM 302, ROM 303, hard disk controller 305, keyboard controller 306, serialization controller 307, parallel controller 308 and display controller 309 arranged.Hard disk 310 and hard disk controller 305 couplings, keyboard 311 and keyboard controller 306 couplings, serial external unit 312 and serial interface controller 307 couplings, parallel external unit 313 and parallel interface controller 308 couplings, and display 314 and display controller 309 couplings.
Should be appreciated that the described structured flowchart of Fig. 3 illustrates just to the purpose of example, rather than limitation of the scope of the invention.In some cases, can increase or reduce as the case may be some equipment.
As mentioned above, equipment 200 can be implemented as pure hardware, such as chip, ASIC, SOC etc.These hardware can be integrated in the computer system 300.In addition, embodiments of the present invention also can realize by the form of computer program.For example, the method 100 of describing with reference to figure 1 can realize by computer program.This computer program can be stored in RAM for example shown in Figure 3 304, ROM 304, hard disk 310 and/or any suitable storage medium, perhaps downloads on the computer system 300 from suitable position by network.Computer program can comprise the computer code part, and it comprises can be by the programmed instruction of suitable treatment facility (for example, the CPU shown in Fig. 3 301) execution.Described programmed instruction can comprise the instruction for the step of implementation method 100 at least.
Above spirit of the present invention and principle have been explained in conjunction with some embodiments.As mentioned above, embodiments of the present invention take full advantage of the characteristics of SNS website.At first, the image in the SNS website is organized take photograph album as unit usually, and the statistical study of SNS website is shown: the user is issue and the same or analogous image of storage class in same photograph album often.In addition, user behavior analysis also shows: often have similar behavior pattern and custom between the good friend of SNS website.Thus, in the bad image detection of SNS website, can also include the SNS incidence relation between the user in consideration.Be attached in the bad image detection by the These characteristics with the SNS website, can effectively improve the Stability and veracity of detection.
Should be noted that embodiments of the present invention can realize by the combination of hardware, software or software and hardware.Hardware components can utilize special logic to realize; Software section can be stored in the storer, and by suitable instruction execution system, for example microprocessor or special designs hardware are carried out.Those having ordinary skill in the art will appreciate that above-mentioned equipment and method can and/or be included in the processor control routine with computer executable instructions realizes, for example such as the mounting medium of disk, CD or DVD-ROM, provide such code such as the programmable memory of ROM (read-only memory) (firmware) or such as the data carrier of optics or electronic signal carrier.Equipment of the present invention and module thereof can be by such as VLSI (very large scale integrated circuit) or gate array, realize such as the semiconductor of logic chip, transistor etc. or such as the hardware circuit of the programmable hardware device of field programmable gate array, programmable logic device etc., also can use the software of being carried out by various types of processors to realize, also can by the combination of above-mentioned hardware circuit and software for example firmware realize.
The communication network of mentioning in the instructions can comprise disparate networks, include but not limited to LAN (Local Area Network) (" LAN "), wide area network (" WAN "), according to the network of IP agreement (for example, the Internet) and ad-hoc network (for example, ad hoc peer-to-peer network).
Although should be noted that some devices or the sub-device of having mentioned equipment in above-detailed, this division only is not enforceable.In fact, according to the embodiment of the present invention, the feature of above-described two or more devices and function can be specialized in a device.Otherwise, the feature of an above-described device and function can Further Division for to be specialized by a plurality of devices.
In addition, although described in the accompanying drawings the operation of the inventive method with particular order,, this is not that requirement or hint must be carried out these operations according to this particular order, or the operation shown in must carrying out all could realize the result of expectation.On the contrary, the step of describing in the process flow diagram can change execution sequence.Additionally or alternatively, can omit some step, a plurality of steps be merged into a step carry out, and/or a step is decomposed into a plurality of steps carries out.
Although described the present invention with reference to some embodiments, should be appreciated that the present invention is not limited to disclosed embodiment.The present invention is intended to contain interior included various modifications and the equivalent arrangements of spirit and scope of claims.The scope of claims meets the most wide in range explanation, thereby comprises all such modifications and equivalent structure and function.
Claims (14)
1. method of using in the bad image detection of social networks SNS website comprises:
To the first image execution analysis in the described SNS website, to determine that described the first image is as the first probability of bad image;
Be in the preset range in response to described the first probability, at least one the second image execution analysis in the photograph album under described the first image, to determine that described at least one second image is as at least one second probability of bad image; And
Regulate described the first probability according to described at least one second probability,
Described the first probability after wherein regulating will be used to judge whether described the first image is bad image.
2. method according to claim 1 further comprises:
Be in the described preset range in response to described the first probability, determine to have the user of described the first image and be bad user's the 3rd probability; And
Regulate described the first probability according to described the 3rd probability.
3. method according to claim 2, wherein said the 3rd probability determine based on following at least one:
Whether described user is in the bad user list;
Whether described user at least one good friend in described SNS website is in the described bad user list; And
The record of the historical behavior of described user in described SNS website.
4. method according to claim 1 further comprises:
Be judged as bad image in response to described the first image according to described the first probability after regulating, increase progressively the counter that is associated with the user who has described the first image, described counter is used for recording described user in the bad behavior of described SNS website,
The value of wherein said counter will be used to judge whether described user is bad user.
5. method according to claim 1 further comprises:
According to described the first probability and described at least one second probability, determine that described photograph album is the 4th probability of bad photograph album.
6. method according to claim 5 further comprises:
Be judged as bad photograph album in response to described photograph album according to described the 4th probability, all spectral discriminations that described photograph album is comprised are bad image.
7. method according to claim 1, wherein to a Given Graph look like to carry out described analysis comprise following at least one:
Analyze the content that described Given Graph picture comprises; And
Analyze the text that looks like to be associated with described Given Graph.
8. equipment that uses in the bad image detection of social networks SNS website comprises:
The first probability is determined device, and configuration is used for the first image execution analysis to described SNS website, to determine that described the first image is as the first probability of bad image;
The second probability is determined device, configuration is used for being in the preset range in response to described the first probability, to at least one the second image execution analysis in the photograph album under described the first image, to determine that described at least one second image is as at least one second probability of bad image; And
The first probability regulating device, configuration are used for regulating described the first probability according to described at least one second probability,
Described the first probability after wherein regulating will be used to judge whether described the first image is bad image.
9. equipment according to claim 8 further comprises:
The 3rd probability is determined device, and configuration is used for being in the described preset range in response to described the first probability, determine to have the user of described the first image and be bad user's the 3rd probability,
Wherein said the first probability regulating device further configuration is used for regulating described the first probability according to described the 3rd probability.
10. equipment according to claim 9, wherein said the 3rd probability determine the device configuration be used for based on following at least one determine described the 3rd probability:
Whether described user is in the bad user list;
Whether described user at least one good friend in described SNS website is in the described bad user list; And
The record of the historical behavior of described user in described SNS website.
11. equipment according to claim 8 further comprises:
The user management device, configuration is used for being judged as bad image in response to described the first image according to described the first probability after regulating and increases progressively the counter that is associated with the user who has described the first image, described counter is used for recording described user in the bad behavior of described SNS website
The value of wherein said counter will be used to judge whether described user is bad user.
12. equipment according to claim 8 further comprises:
The 4th probability is determined device, and configuration is used for according to described the first probability and described at least one second probability, determines that described photograph album is the 4th probability of bad photograph album.
13. equipment according to claim 12 further comprises:
Photograph album management devices, configuration are used for being judged as bad photograph album in response to described photograph album according to described the 4th probability, and all spectral discriminations that described photograph album is comprised are bad image.
14. equipment according to claim 8, wherein to a Given Graph look like to carry out described analysis comprise following at least one:
Analyze the content that described Given Graph picture comprises; And
Analyze the text that looks like to be associated with described Given Graph.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110323761.7A CN103064858B (en) | 2011-10-19 | 2011-10-19 | Method and apparatus for objectionable image detection in social networking websites |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110323761.7A CN103064858B (en) | 2011-10-19 | 2011-10-19 | Method and apparatus for objectionable image detection in social networking websites |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103064858A true CN103064858A (en) | 2013-04-24 |
| CN103064858B CN103064858B (en) | 2016-03-30 |
Family
ID=48107488
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110323761.7A Active CN103064858B (en) | 2011-10-19 | 2011-10-19 | Method and apparatus for objectionable image detection in social networking websites |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103064858B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105703966A (en) * | 2014-11-27 | 2016-06-22 | 阿里巴巴集团控股有限公司 | Internet behavior risk identification method and apparatus |
| CN106101740A (en) * | 2016-07-13 | 2016-11-09 | 百度在线网络技术（北京）有限公司 | A kind of video content recognition method and apparatus |
| CN107729924A (en) * | 2017-09-25 | 2018-02-23 | 平安科技（深圳）有限公司 | Picture review probability interval generation method and picture review decision method |
| CN108595422A (en) * | 2018-04-13 | 2018-09-28 | 卓望信息技术（北京）有限公司 | A method of the bad multimedia message of filtering |
| CN112651352A (en) * | 2020-12-30 | 2021-04-13 | 深圳市商汤科技有限公司 | Image processing method and device, electronic equipment and storage medium |
| WO2021069989A1 (en) * | 2019-10-06 | 2021-04-15 | International Business Machines Corporation | Filtering group messages |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1396532A (en) * | 2001-07-16 | 2003-02-12 | 友立资讯股份有限公司 | Sexy picture checking system and method |
| US20050160258A1 (en) * | 2003-12-11 | 2005-07-21 | Bioobservation Systems Limited | Detecting objectionable content in displayed images |
| KR20100034782A (en) * | 2008-09-25 | 2010-04-02 | 고려대학교 산학협력단 | System and method for detecting pornographic contents |
| CN102208992A (en) * | 2010-06-13 | 2011-10-05 | 天津海量信息技术有限公司 | Internet-facing filtration system of unhealthy information and method thereof |
-
2011
- 2011-10-19 CN CN201110323761.7A patent/CN103064858B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1396532A (en) * | 2001-07-16 | 2003-02-12 | 友立资讯股份有限公司 | Sexy picture checking system and method |
| US20050160258A1 (en) * | 2003-12-11 | 2005-07-21 | Bioobservation Systems Limited | Detecting objectionable content in displayed images |
| KR20100034782A (en) * | 2008-09-25 | 2010-04-02 | 고려대학교 산학협력단 | System and method for detecting pornographic contents |
| CN102208992A (en) * | 2010-06-13 | 2011-10-05 | 天津海量信息技术有限公司 | Internet-facing filtration system of unhealthy information and method thereof |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105703966A (en) * | 2014-11-27 | 2016-06-22 | 阿里巴巴集团控股有限公司 | Internet behavior risk identification method and apparatus |
| CN106101740A (en) * | 2016-07-13 | 2016-11-09 | 百度在线网络技术（北京）有限公司 | A kind of video content recognition method and apparatus |
| CN106101740B (en) * | 2016-07-13 | 2019-12-24 | 百度在线网络技术（北京）有限公司 | Video content identification method and device |
| CN107729924A (en) * | 2017-09-25 | 2018-02-23 | 平安科技（深圳）有限公司 | Picture review probability interval generation method and picture review decision method |
| CN108595422A (en) * | 2018-04-13 | 2018-09-28 | 卓望信息技术（北京）有限公司 | A method of the bad multimedia message of filtering |
| WO2021069989A1 (en) * | 2019-10-06 | 2021-04-15 | International Business Machines Corporation | Filtering group messages |
| CN114514730A (en) * | 2019-10-06 | 2022-05-17 | 国际商业机器公司 | Filtering group messages |
| GB2604772A (en) * | 2019-10-06 | 2022-09-14 | Ibm | Filtering group messages |
| US11552914B2 (en) | 2019-10-06 | 2023-01-10 | International Business Machines Corporation | Filtering group messages |
| CN112651352A (en) * | 2020-12-30 | 2021-04-13 | 深圳市商汤科技有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN112651352B (en) * | 2020-12-30 | 2022-07-19 | 深圳市商汤科技有限公司 | Image processing method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103064858B (en) | 2016-03-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10489830B2 (en) | Aggregation of rating indicators | |
| CN103064858B (en) | Method and apparatus for objectionable image detection in social networking websites | |
| US9338047B1 (en) | Detecting content on a social network using browsing patterns | |
| Pesce et al. | Privacy attacks in social media using photo tagging networks: a case study with Facebook | |
| CN106682906B (en) | Risk identification and service processing method and equipment | |
| CN106874253A (en) | Recognize the method and device of sensitive information | |
| Paul et al. | Elites tweet? Characterizing the Twitter verified user network | |
| CN104809132A (en) | Method and device for acquiring social relation type of network subject | |
| Kim et al. | Objectivity and subjectivity in aesthetic quality assessment of digital photographs | |
| CN110532480B (en) | Knowledge graph construction method for recommending human-read threat information and threat information recommendation method | |
| KR102083624B1 (en) | System and method for analyzing interest object, and apparatus applied to the same | |
| Şensoy et al. | Stage: Stereotypical trust assessment through graph extraction | |
| WO2020048056A1 (en) | Risk decision method and apparatus | |
| CN110362601A (en) | Mapping method, device, equipment and the storage medium of metadata standard | |
| WO2019242627A1 (en) | Data processing method and apparatus | |
| CN106779271A (en) | Determine the method and device and account risk score method and device of feature of risk | |
| CN110046251A (en) | Community content methods of risk assessment and device | |
| Mashhadi et al. | No walk in the park: The viability and fairness of social media analysis for parks and recreational policy making | |
| KR102135074B1 (en) | System for identifying fake news using artificial-intelligence-based fact-checking guidelines | |
| US20220004652A1 (en) | Providing images with privacy label | |
| Jiang et al. | Camera fingerprint: A new perspective for identifying user's identity | |
| Li et al. | Social network-based recommendation: a graph random walk kernel approach | |
| CN103353868B (en) | A kind of method and apparatus for the resource assessment information for being used to determine multimedia resource | |
| Chaudhry et al. | Spam detection in social network using machine learning approach | |
| CN116860963A (en) | Text classification method, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20200313 Address after: Room 307, 3 / F, Xinghua green industrial building, Baosheng East Road, Haidian District, Beijing (Dongsheng District) Patentee after: Beijing matador culture media Co., Ltd Address before: 100041, B1011 building, North building, 11 Xing Dong Street, Shijingshan District, Beijing Patentee before: BEIJING OAK PACIFIC NETSCAPE TECHNOLOGY DEVELOPMENT Co.,Ltd. |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20220429 Address after: 100000 duoniu media building, Baosheng East Road, Haidian District, Beijing Patentee after: Chengdu Renren mutual Entertainment Technology Co.,Ltd. Address before: 100192 room 307, 3 / F, Xinghua green industry building, Baosheng East Road, Haidian District, Beijing (Dongsheng District) Patentee before: Beijing matador culture media Co.,Ltd. |
|
| TR01 | Transfer of patent right |