WO2015081748A1 - 图片内容属性识别方法和系统 - Google Patents
图片内容属性识别方法和系统 Download PDFInfo
- Publication number
- WO2015081748A1 WO2015081748A1 PCT/CN2014/087109 CN2014087109W WO2015081748A1 WO 2015081748 A1 WO2015081748 A1 WO 2015081748A1 CN 2014087109 W CN2014087109 W CN 2014087109W WO 2015081748 A1 WO2015081748 A1 WO 2015081748A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- picture
- homologous
- cluster
- pictures
- content attribute
- 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.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
Definitions
- the present invention relates to the field of image recognition, and in particular to a method and system for identifying a picture content attribute.
- advertisement images which are very rich in types, including advertisements for various commodities (for example, advertisements for milk powder and clothes), and advertisements for physical stores, and some Other types of ads.
- ad images will appear not only on the merchant's site, but also on other resource sites, for example, in communities that allow users to upload images ( forums, photo sites, etc.), some users will upload ad images.
- the existence of a large number of advertisement images often causes interference to users, and even when users perform image search, advertisement images that are not related to user needs may appear.
- the present invention has been made in order to provide a picture content attribute recognition method and system that overcomes the above problems or at least partially solves the above problems.
- a picture content attribute identifying method including: calculating a relative number of reloads of a plurality of homologous picture clusters for a specific resource station; and identifying corresponding homologous picture clusters according to the relative reload number Image content properties.
- the method before calculating the relative number of reloads of the plurality of homologous picture clusters for the specific resource site, the method further includes: performing similar picture recognition on the collected pictures, and aggregating the pictures into multiple homologous picture clusters.
- identifying, according to the relative number of reloads, a picture content attribute in the corresponding homologous picture cluster including:
- the picture content attribute in the target picture cluster is identified according to the trained filter model.
- a picture content attribute recognition system comprising: a relative load number calculation module, configured to calculate a relative number of reloads of a plurality of homologous picture clusters for a specific resource site; and a picture content attribute recognition module And for identifying a picture content attribute in the corresponding homologous picture cluster according to the relative number of reloads.
- a computer program comprising computer readable code, when said computer readable code is run on a computing device, causing said computing device to perform any of claims 1-12 A method for identifying a picture content attribute.
- a computer readable medium storing the computer program according to claim 23 is provided.
- the picture content attribute identification method and system utilizes the relative retransmission number of the homologous picture cluster to identify the picture content attribute in the corresponding homologous picture cluster, and the data used for the picture attribute identification is relatively reproduced.
- the number because the relative number of reloads is the data that reflects the proportion of the station outside the station on the specific resource site, and a main feature of the image as an advertisement is that the number of reprints on a resource site is very high, while on the Internet.
- the number of reprints on other resource sites in the scope will be relatively less, so the relative reload number can be used to distinguish whether the image is propagated as an advertisement, and the image content attribute of the image can be accurately identified by using the relative number of reloads. Determine if the image is an ad image.
- the picture cluster is a unit for identifying the image content attribute to determine whether the picture in each of the homologous picture clusters is an advertisement picture.
- the training of the filter model is performed by using the relative retransmission number of the homologous picture cluster for the specific resource site as the training data, and the filter model using the relative reload number is trained, and the obtained filter model can directly take pictures of the picture.
- the content attribute is identified to accurately determine whether the picture is an advertisement picture.
- FIG. 1 shows a flow chart of a picture content attribute recognition method according to an embodiment of the present invention
- FIG. 2 shows a block diagram of a picture content attribute recognition system in accordance with one embodiment of the present invention
- FIG. 3 is a flow chart showing a method for identifying a picture content attribute according to an embodiment of the present invention
- FIG. 4 shows a partial flow chart of a picture content attribute identification method in accordance with one embodiment of the present invention
- FIG. 5 is a flow chart showing a method for identifying a picture content attribute according to an embodiment of the present invention
- Figure 6 shows a block diagram of a picture content attribute recognition system in accordance with one embodiment of the present invention
- Figure 7 shows a block diagram of a picture content attribute recognition system in accordance with one embodiment of the present invention.
- Figure 8 shows a block diagram of a picture content attribute recognition system in accordance with one embodiment of the present invention.
- FIG. 9 is a flowchart showing a picture content attribute identifying method according to an embodiment of the present invention.
- FIG. 10 is a partial flow chart showing a method for identifying a picture content attribute according to an embodiment of the present invention.
- FIG. 11 is a flowchart showing a picture content attribute recognizing method according to an embodiment of the present invention.
- Figure 12 shows a block diagram of a picture content attribute recognition system in accordance with one embodiment of the present invention.
- Figure 13 shows a block diagram of a picture content attribute recognition system in accordance with one embodiment of the present invention.
- Figure 14 shows a block diagram of a picture content attribute recognition system in accordance with one embodiment of the present invention.
- Figure 15 is a block diagram schematically showing a computing device for performing a picture content attribute identifying method in accordance with the present invention
- Fig. 16 schematically shows a storage unit for holding or carrying program code implementing the picture content attribute recognizing method according to the present invention.
- FIG. 1 shows a process flow diagram of a picture content attribute identifying method in accordance with one embodiment of the present invention. Referring to Figure 1, the method includes:
- Step S102 Calculate a relative number of retransmissions of a plurality of homologous picture clusters for a specific resource station
- Step S104 Identify a picture content attribute in the corresponding homologous picture cluster according to the relative number of retransmissions.
- FIG. 2 is a block diagram showing the structure of a picture content attribute recognition system according to an embodiment of the present invention. Referring to Figure 2, the system includes:
- the relative reload number calculation module 210 is configured to calculate a relative retransmission number of a plurality of homologous picture clusters for a specific resource station;
- the picture content attribute identification module 220 is coupled to the relative reload number calculation module 210 for identifying the picture content attribute in the corresponding homologous picture cluster according to the relative reload number.
- the flow shown in FIG. 1 indicates that the image content attribute can be identified according to the relative number of reloads, but there are different means or methods for how to implement this function. Different implementation manners are described in detail in different embodiments.
- an embodiment of the present invention provides a method for identifying a content attribute of a picture, including: Step 310: Calculating a relative number of reloads of a plurality of homologous picture clusters for a specific resource site, each A picture cluster is an aggregation of a group of pictures, for example, a group of pictures with a higher degree of similarity, and a relative number of retransmissions is a retransmission ratio of a picture that can reflect a picture of a homologous picture group outside a station of a specific resource station.
- the data, the relative number of reloads is calculated in many ways.
- the calculation method of the relative reload number is not limited; in step 320, the filter model is trained according to the plurality of homologous picture clusters and the corresponding relative reload number, and the advertisement picture is adopted.
- the advertisement picture has the following characteristics: the production cost of the advertisement picture is high, and many advertisement pictures are made by the merchants who spend money and spend time. Because the production cost of the advertisement picture is high, the merchant will spread an advertisement picture many times. However, these advertisement images are basically only spread by merchants, while other users basically do not spread advertisement images.
- the training filter model Tools include but not The method is limited to the open source LIB SVM; in step 330, the image content attribute in the target image cluster is identified according to the trained filter model, that is, whether the image in the target image cluster is an advertisement image, which is beneficial to filtering the advertisement image and the like, thereby avoiding The advertisement image affects the user's experience.
- the advertisement image can be identified and filtered, thereby using the non-advertisement image as a search.
- the results are provided to the user to ensure the user's experience.
- other features are also considered, such as the length/width of the picture, the size of the picture, the clarity of the picture, whether the picture link is co-located with the web page, or the picture. Whether the jump link is outside the station, etc., in the training filter, according to the relative retransmission number of each of the homologous picture clusters, and the length/width of the picture in the picture cluster, the size of the picture, the clarity of the picture, the picture Whether the link is on the same page as the web page, whether the picture jump link is one or more combinations in the station, first through the filter to learn and train.
- the target picture cluster is identified, it is also filtered and identified as an advertisement picture by referring to one or more of these other features.
- the picture content attribute identification method of the embodiment may include: for one of a plurality of homologous picture clusters Picture cluster, the number of reprints of pictures in a cluster of homologous pictures on a particular resource site, for example, 30 times on picture station A, compared to the number of reprints on multiple resource sites, for example at 10 picture stations (including picture station A) is reprinted 35 times, and the relative retransmission number of the homologous picture cluster for a specific resource site is obtained.
- the multiple resource sites include a specific resource site.
- a feasible way of calculating the relative reload number is provided.
- the specific comparison method is not limited. For example, it is possible to take 30/35, 30/(35-30) as the relative number of reloads.
- step 310 includes: step 311, calculating a specific resource site.
- the first average number of reprints of the picture above for example, assume that the first average number of reprints of picture station A is 5; step 312, calculating the second average number of retransmissions of pictures on multiple resource sites, for example, assuming 10 picture stations (including The second average number of reprints of the picture station A) is 20; in step 313, the first difference between the number of reprints of the picture in the same-origin picture group and the first average number of retransmissions is taken, and the first difference is actually The difference between the picture of the homologous picture cluster and the other pictures on the specific resource site may be reflected.
- the first difference value and the second difference value comparison manner are not limited, for example, 25/15, (25 ⁇ a) /(15 ⁇ b) is ok, a and b are constants.
- a method for obtaining the average number of retransmissions quickly and efficiently is provided.
- step 301 grasping Taking a picture link (URL) appearing on multiple resource sites; step 302, detecting whether the link corresponding to the picture link and the picture of the same-origin picture group is the same, which reflects whether a picture is reproduced with a different URL, and/or Detecting whether the check information of the picture corresponding to the picture link and the check information of the picture of the same picture cluster (including but not limited to the MD5 value) are the same, which reflects whether there are multiple identical pictures, and/or the corresponding picture link is detected.
- step 301 grasping Taking a picture link (URL) appearing on multiple resource sites
- step 302 detecting whether the link corresponding to the picture link and the picture of the same-origin picture group is the same, which reflects whether a picture is reproduced with a different URL, and/or Detecting whether the check information of the picture corresponding to the picture link and the check information of the picture of the same picture cluster (including but not limited to the MD5 value) are the same, which reflects whether there are
- step 303 determining whether the image link is a reprint of the image of the homologous picture cluster, and counting the homologous picture For the number of reprints of the pictures of the cluster, in this embodiment, a technical solution for comprehensively counting the number of picture reloads is provided.
- Another embodiment of the present invention provides a picture content attribute identification method.
- the picture content attribute identification method in the embodiment the specific resource site reproduces each homologous picture in a plurality of homologous picture clusters.
- the site with the most pictures of the cluster, the site with the most number of reprinted images is likely to be the site for the merchants of the advertisement image, and the number of reprints corresponding to the site can most effectively reflect whether the image is an advertisement image.
- Another embodiment of the present invention provides a method for identifying a picture content attribute.
- the picture content attribute recognition method of the embodiment corresponds to the same source picture, and each of the same The picture of the source picture cluster has the same image feature as the corresponding source picture.
- the picture of each of the same picture clusters is the same, or can be modified by the same picture.
- Image features in the image include, but are not limited to, contour features, color features, histogram features, and the like.
- an embodiment of the present invention provides a picture content attribute recognition system for supporting the picture content attribute identification method of the first embodiment, and includes other modules in addition to the modules in FIG. Specifically, it includes: a relative load number calculation module 210, configured to calculate a relative number of reloads of a plurality of homologous picture clusters for a specific resource site, where each picture cluster is an aggregation of a group of pictures, for example, may be similarity A high set of pictures, and the relative number of retransmissions is a kind of data that can reflect the retransmission ratio of the pictures of the homologous picture clusters in the station of the specific resource station.
- a relative load number calculation module 210 configured to calculate a relative number of reloads of a plurality of homologous picture clusters for a specific resource site, where each picture cluster is an aggregation of a group of pictures, for example, may be similarity A high set of pictures, and the relative number of retransmissions is a kind of data that can reflect the
- the relative retransmission number is calculated in many ways, and the relative retransmission is not in this embodiment.
- the calculation method of the number is limited; the training module 230 is configured to input a plurality of homologous picture clusters and corresponding relative reload numbers into the training filter model in the filter.
- the advertisement picture has the following characteristics: the production cost of the advertisement picture is high, and many advertisement pictures are It is the merchant who spends money and spends time making it. Because the production cost of the advertisement image is high, the merchant will spread an advertisement image many times, but these advertisement images are basically only spread by the merchants, while other users basically do not.
- the filter 240 is adapted to obtain the trained filter model according to the training module, and filter the target image cluster according to the model.
- the filter used in this embodiment includes but is not limited to the open source LIB SVM; the picture content attribute identification module 220 And filtering the target image cluster according to the filter 240 to identify the image in the target image cluster.
- Content attributes that identify the target picture cluster picture is a picture ads.
- the system further includes: a picture format feature module and/or a picture link feature module; the picture format feature module is adapted to extract a homology picture cluster and a format feature of a picture included in the target picture cluster;
- the picture link feature module is adapted to extract a homology picture cluster and a link feature of a picture included in the target picture cluster;
- the training module 230 is further adapted to be based on a plurality of homologous picture clusters, corresponding relative reload numbers, and corresponding
- the picture format feature and/or the picture link feature are input together with the training filter model in the filter; the filter 240 is further adapted to combine the relative number of reloads corresponding to the target picture cluster and the corresponding picture format feature according to the trained model.
- the target picture cluster is filtered;
- the picture content attribute identification module 220 is further configured to: according to the relative number of reloads corresponding to the target picture cluster and the corresponding picture format feature and/or picture according to the filter
- the link feature filters the target image cluster to identify the image in the target image cluster Capacity attribute.
- the technical solution of the embodiment can identify the advertisement image from the image. And filtering to provide non-advertising images as search results to users, thus ensuring the user experience.
- other features are also considered, such as the length/width of the picture, the size of the picture, the clarity of the picture, whether the picture link is co-located with the web page, or the picture is skipped. Whether the link is a station or not, it is also learned and trained through the classifier. When the target picture cluster is identified, one or more of these other features are also considered for screening and identifying whether it is an advertisement picture.
- the picture content attribute recognition system of the embodiment has a relative load number calculation module 210 for one of a plurality of homologous picture clusters.
- the source picture cluster the number of reprints of pictures in the same-origin picture cluster on a specific resource site, for example, reprinted 30 times on picture station A, compared with the number of reprints on multiple resource stations, for example, in 10 pictures
- the station including picture station A
- the station is reprinted 35 times, and the relative number of reloads of the homologous picture cluster for a specific resource site is obtained.
- the multiple resource sites include a specific resource site.
- a feasible way of calculating the relative reload number is provided.
- the specific comparison method is not limited. For example, it is possible to take 30/35, 30/(35-30) as the relative number of reloads.
- the picture content attribute recognition system of the embodiment further includes: a first average number of reloads calculation module 250. For Calculating a first average number of reprints of the picture on the specific resource site, for example, assuming that the first average number of reprints of the picture station A is 5; and a second average number of retransmissions calculation module 260 for calculating the second picture of the plurality of resource sites
- the average number of reloads for example, assumes that the second average number of reprints of 10 picture stations (including picture station A) is 20; the relative number of retransmissions calculation module 210 takes the number of pictures of the pictures in the same-origin picture group on the specific resource site and the first
- the first difference of the average number of reprints, the first difference may actually reflect the difference of the retransmission of the picture of the homologous picture cluster with other pictures on a specific resource site, and
- the second difference between the number of reprints of the pictures in the same-origin picture group at the multiple resource sites and the second average number of retransmissions is taken.
- the value, the second difference may actually reflect the difference in the reprinting of the picture of the same-origin picture cluster with other pictures on multiple resource sites, and the greater the difference, the less likely the homologous picture cluster is the advertising picture, combined with the foregoing Implementation
- the first difference value and the second difference value are compared to obtain the relative number of reloads of the homologous picture cluster for a specific resource station.
- another calculated relative reload number is provided.
- the first difference and the second difference are not compared. For example, it is possible to take 25/15, (25 ⁇ a)/(15 ⁇ b), and a and b are constants.
- the first average number of reloads calculation module 250 takes a picture of a plurality of homologous picture clusters.
- the plurality of pictures located on a specific resource site compares the number of the plurality of pictures with the number of the same picture clusters corresponding to the plurality of pictures to obtain a first average number of reprints, for example, 100 pictures on the picture station A,
- the technical solution in this embodiment provides a fast and efficient way to obtain the average number of retransmissions.
- the picture content attribute recognition system of the embodiment further includes: a picture link capture module 270,
- the image link detection module 280 is configured to detect whether the link corresponding to the image of the image link and the image of the same-origin picture cluster is the same, which reflects whether a picture is different.
- the URL is reprinted, and/or whether the check information of the picture corresponding to the picture link and the check information of the picture of the same picture cluster (including but not limited to the MD5 value) are the same, which reflects whether there are multiple identical pictures.
- the feature includes, but is not limited to, a contour feature, a color feature, a histogram feature, and the like; a picture reload number statistics module 290, configured to determine, according to the detection result, whether the image link is For the retransmission of the pictures of the homologous picture clusters, and counting the number of reprints of the pictures of the homologous picture clusters, the present embodiment provides a technical solution for comprehensively counting the picture retransmission numbers.
- Another embodiment of the present invention provides a picture content attribute recognition system.
- the specific resource site reproduces each homologous picture in a plurality of homologous picture clusters.
- the site with the most pictures of the cluster, the site with the most number of reprinted images is likely to be the site for the merchants of the advertisement image, and the number of reprints corresponding to the site can most effectively reflect whether the image is an advertisement image.
- Another embodiment of the present invention provides a picture content attribute recognition system.
- the picture of each homologous picture cluster corresponds to the same source picture, and each of the same The picture of the source picture cluster has the same image feature as the corresponding source picture.
- the picture of each of the same picture clusters is the same, or can be modified by the same picture.
- Image features in the image include, but are not limited to, contour features, color features, histogram features, and the like.
- Step 910 Perform similar picture recognition on the collected picture, and aggregate the picture into multiple homologous picture clusters.
- similar pictures are aggregated into the same homologous picture cluster.
- the remaining pictures are also necessarily advertisement pictures, so in this embodiment
- the image content attribute is identified by the picture cluster as a unit to determine whether the picture in each of the homologous picture clusters is an advertisement picture, and the similar picture can be identified based on the current image recognition technology. This embodiment does not limit the recognition technology of similar pictures.
- Step 920 Calculate a relative number of reloads of a plurality of homologous picture clusters for a specific resource station, and the relative retransmission number is data that reflects a retransmission ratio of pictures of the same-origin picture group outside the station of the specific resource station, and the relative number of reloads
- the calculation method of the relative number of reloads is not limited; in step 930, according to the relative rotation The load number identifies the image content attribute in the corresponding homologous picture cluster.
- the advertisement picture has the following characteristics: the production cost of the advertisement picture is high, and many advertisement pictures are made by the merchants who spend money and spend time, because The production cost of the advertisement image is high, so the merchant will spread an advertisement image many times, but these advertisement images are basically only spread by the merchant, while other users basically do not spread the advertisement image, and the advertisement image is spread.
- the difference will eventually be reflected in the number of reprints on the resource site: the number of reprints on a particular resource site is very high (the merchant deliberately spreads), while the number of reposts on other sites on the Internet is relatively small (other users are not Propagation), that is, the reprinting ratio of the advertisement picture outside the station in the specific resource site station is relatively high, so the relative retransmission number can be used as a kind of data for distinguishing the advertisement picture from the non-advertisement picture, so the technical solution of the embodiment can identify the homologous Whether the image in the image cluster is an advertisement image, which is useful for filtering the advertisement image, etc.
- the advertisement picture can be identified and filtered, thereby The advertisement image is provided to the user as a search result, thereby ensuring the user's experience.
- the SVM model may also be used to learn and train, and the relative number of reloads and the combination of one or more of the other features are classified as parameter pairs. The device performs training, and also uses the trained SVM model and corresponding features as parameters to identify at the time of final recognition.
- the picture content attribute identification method of the embodiment may include: performing homology to one of the plurality of homologous picture clusters.
- Picture cluster the number of reprints of pictures in a cluster of homologous pictures on a particular resource site, for example, 30 times on picture station A, compared to the number of reprints on multiple resource sites, for example at 10 picture stations (including picture station A) is reprinted 35 times, and the relative retransmission number of the homologous picture cluster for a specific resource site is obtained.
- the multiple resource sites include a specific resource site.
- a feasible way of calculating the relative reload number is provided.
- the specific comparison method is not limited. For example, it is possible to take 30/35, 30/(35-30) as the relative number of reloads.
- step 920 includes: step 921, calculating a specific resource site.
- the first average number of reprints of the picture above for example, assume that the first average number of reprints of picture station A is 5; step 922, calculating the second average number of retransmissions of pictures on multiple resource sites, for example, assuming 10 picture stations (including The second average number of reprints of the picture station A) is 20; in step 923, the first difference between the number of reprints of the picture in the same-origin picture group and the first average number of retransmissions is taken, and the first difference is actually The difference between the picture of the homologous picture cluster and the other pictures on the specific resource site may be reflected.
- the first difference value and the second difference value comparison manner are not limited, for example, 25/15, (25 ⁇ a) /(15 ⁇ b) is ok, a and b are constants.
- a method for obtaining the average number of retransmissions quickly and efficiently is provided.
- another embodiment of the present invention provides a method for identifying a picture content attribute.
- the method for identifying a picture content attribute of the present embodiment before step 920, further includes: step 911, grasping Take multiple a URL appearing on the resource site; step 912, detecting whether the link corresponding to the picture link and the picture of the homologous picture cluster is the same, which reflects whether a picture is reprinted with a different URL, and/or detecting a picture corresponding to the picture link.
- step 913 according to the detection result, determining whether the picture link is a reprint of the picture of the same-origin picture cluster, and counting the number of reprints of the picture of the same-origin picture group, in this embodiment, a comprehensive statistical picture reprint number is provided.
- the technical solution wherein the order of step 911 and step 910 is not limited.
- Another embodiment of the present invention provides a picture content attribute identification method.
- the picture content attribute identification method in the embodiment the specific resource site reproduces each homologous picture in a plurality of homologous picture clusters.
- the site with the most pictures of the cluster, the site with the most number of reprinted images is likely to be the site for the merchants of the advertisement image, and the number of reprints corresponding to the site can most effectively reflect whether the image is an advertisement image.
- Another embodiment of the present invention provides a method for identifying a picture content attribute.
- the picture content attribute recognition method of the embodiment corresponds to the same source picture, and each of the same The picture of the source picture cluster has the same image feature as the corresponding source picture.
- the picture of each of the same picture clusters is the same, or can be modified by the same picture.
- Image features in the image include, but are not limited to, contour features, color features, histogram features, and the like.
- an embodiment of the present invention provides a picture content attribute recognition system for supporting the picture content attribute identification method provided in the second embodiment, and includes other modules in addition to the modules in FIG.
- the image aggregation module 1210 is configured to perform similar image recognition on the collected pictures, and aggregate the pictures into multiple homologous picture clusters.
- similar pictures are aggregated into the same homologous picture cluster.
- the image content attributes are identified by the picture cluster as a unit to determine each Whether the picture in the source picture cluster is an advertisement picture, and the similar picture can be identified based on the current image recognition technology.
- the relative load number calculation module 210 is used to calculate multiple homologous picture clusters.
- the relative retransmission number is a kind of data that can reflect the retransmission ratio of the pictures of the homologous picture clusters in the station of the specific resource station.
- the relative retransmission number is calculated in many ways, which is not correct in this embodiment.
- the method for calculating the relative number of reloads is limited; the picture content attribute identifying module 220 is configured to recognize the number of relative reloads The image content attribute in the corresponding homologous picture cluster.
- the advertisement picture has the following characteristics: the production cost of the advertisement picture is high, and many advertisement pictures are made by the merchants who spend money and spend time. Because the production cost of the advertisement picture is high, the merchant will take an advertisement picture. Spread many times, but these advertisement images are basically only spread by merchants, while other users basically do not spread advertisement images. The difference in the spread of advertisement images will eventually be reflected in the number of reprints on the resource site: The number of reposts on a particular resource site is very high (the merchant deliberately spreads), while the number of reposts on other sites on the Internet is relatively small (other users do not spread), that is, the advertisement image is outside the station of the specific resource site.
- the reprinting ratio will be higher, so the relative reloading number can be used as a distinction between advertising images and non- A kind of data of the advertisement picture, so the technical solution of the embodiment can identify whether the picture in the same-origin picture group is an advertisement picture, which is beneficial to filtering the advertisement picture and the like, so as to prevent the advertisement picture from affecting the user experience, and assume the same
- the source picture cluster is a group of pictures corresponding to the image search request.
- the advertisement picture can be identified and filtered, so that the non-advertisement picture is provided as a search result to the user, thereby ensuring the user's use. Experience.
- the system further includes: a picture format feature module and/or a picture link feature module; the picture format feature module is adapted to extract a format feature of a picture included in the same-origin picture group; the picture link feature The module is adapted to extract a link feature of the picture included in the cluster of the same-origin picture; the picture content attribute identification module 220 is further adapted to: according to the relative number of reprints corresponding to the same-origin picture group and the corresponding picture format feature and/or picture link feature pair The homologous picture clusters are screened to identify the image content attributes in the homologous picture cluster.
- the SVM model may also be used to learn and train, and the combination of the relative number of reloads and one or more of the other features may be used as a parameter to train the classifier, and at the time of final recognition.
- the trained SVM model is also used to identify.
- the picture content attribute recognition system of the embodiment has a relative load number calculation module 210 for one of a plurality of homologous picture clusters.
- the source picture cluster the number of reprints of pictures in the same-origin picture cluster on a specific resource site, for example, reprinted 30 times on picture station A, compared with the number of reprints on multiple resource stations, for example, in 10 pictures
- the station including picture station A
- the station is reprinted 35 times, and the relative number of reloads of the homologous picture cluster for a specific resource site is obtained.
- the multiple resource sites include a specific resource site.
- a feasible way of calculating the relative reload number is provided.
- the specific comparison method is not limited. For example, it is possible to take 30/35, 30/(35-30) as the relative number of reloads.
- the picture content attribute recognition system of the embodiment further includes: a first average number of reloads calculation module 1240. For calculating a first average number of reloads of pictures on a specific resource site, for example, assuming that the first average number of reprints of the picture station A is 5; and a second average number of retransmissions calculation module 1250 for calculating pictures on the plurality of resource sites
- the second average number of reprints for example, assumes that the second average number of reprints of 10 picture stations (including picture station A) is 20; the relative number of retransmissions calculation module 210 takes the number of pictures of the pictures in the same-origin picture group at a specific resource site.
- the second difference the second difference actually reflects the difference between the pictures of the homologous picture cluster and the other pictures on multiple resource sites. The greater the difference, the more likely the homologous picture cluster is the advertising picture.
- the first difference and the second difference are not limited, for example, 25/15, ( 25 ⁇ a)/(15 ⁇ b) are all possible, and a and b are constants.
- the first average number of reloads calculation module 240 takes a picture of a plurality of homologous picture clusters.
- the plurality of pictures located on a specific resource site compares the number of the plurality of pictures with the number of the same picture clusters corresponding to the plurality of pictures to obtain a first average number of reprints, for example, 100 pictures on the picture station A,
- the technical solution in this embodiment provides a fast and efficient way to obtain the average number of retransmissions.
- the picture content attribute recognition system of the embodiment further includes: a picture link capture module 1260,
- the URL link detection module 1270 is configured to detect whether the link corresponding to the picture link and the picture of the same-origin picture group is the same, which reflects whether a picture is reprinted with a different URL.
- the picture reload number statistics module 1280 is configured to determine whether the image link is a homologous map according to the detection result. Reprinted number of clusters reproduced picture, and the picture statistics homologous cluster of images, the present embodiment provides a comprehensive statistical picture can be reproduced in a number of technical solutions.
- Another embodiment of the present invention provides a picture content attribute recognition system.
- the specific resource site reproduces each homologous picture in a plurality of homologous picture clusters.
- the site with the most pictures of the cluster, the site with the most number of reprinted images is likely to be the site for the merchants of the advertisement image, and the number of reprints corresponding to the site can most effectively reflect whether the image is an advertisement image.
- Another embodiment of the present invention provides a picture content attribute recognition system.
- the picture of each homologous picture cluster corresponds to the same source picture, and each of the same The picture of the source picture cluster has the same image feature as the corresponding source picture.
- the picture of each of the same picture clusters is the same, or can be modified by the same picture.
- Image features in the image include, but are not limited to, contour features, color features, histogram features, and the like.
- modules in the devices of the embodiments can be adaptively changed and placed in one or more devices different from the embodiment.
- the modules or units or components of the embodiments may be combined into one module or unit or component, and further they may be divided into a plurality of sub-modules or sub-units or sub-components.
- any combination of the features disclosed in the specification, including the accompanying claims, the abstract and the drawings, and any methods so disclosed, or All processes or units of the device are combined.
- Each feature disclosed in this specification (including the accompanying claims, the abstract and the drawings) may be replaced by alternative features that provide the same, equivalent or similar purpose.
- the various component embodiments of the present invention may be implemented in hardware, or in a software module running on one or more processors, or in a combination thereof.
- a microprocessor or digital signal processor may be used in practice to implement some or all of the functionality of some or all of the components of the picture content attribute recognition system in accordance with embodiments of the present invention.
- the invention can also be implemented as a device or device program (e.g., a computer program and a computer program product) for performing some or all of the methods described herein.
- a program implementing the invention may be stored on a computer readable medium or may be in the form of one or more signals. Such signals may be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.
- Figure 15 illustrates a computing device that can implement a picture content attribute recognition method in accordance with the present invention.
- the computing device conventionally includes a processor 1510 and a computer program product or computer readable medium in the form of a memory 1520.
- the memory 1520 may be an electronic memory such as a flash memory, an EEPROM (Electrically Erasable Programmable Read Only Memory), an EPROM, a hard disk, or a ROM.
- Memory 1520 has a storage space 1530 for program code 1531 for performing any of the method steps described above.
- storage space 1530 for program code may include various program code 1531 for implementing various steps in the above methods, respectively.
- the program code can be read from or written to one or more computer program products.
- the computer program product includes program code carriers such as hard disks, compact disks (CDs), memory cards or floppy disks.
- the computer program product is typically a portable or fixed storage unit as described with reference to FIG.
- the storage unit may have a storage segment, a storage space, and the like that are similarly arranged to the storage 1520 in the computing device of FIG.
- the program code can be compressed, for example, in an appropriate form.
- the storage unit includes computer readable code 1531', ie, code that can be read by, for example, a processor such as 1510, which when executed by the computing device causes the computing device to perform each of the methods described above step.
Landscapes
- Engineering & Computer Science (AREA)
- Library & Information Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
一种图片内容属性识别方法和系统,方法包括:计算多个同源图片簇对于特定资源站点的相对转载数;根据所述相对转载数训练识别对应的同源图片簇中的图片内容属性。优点在于,根据图片在网络上被转载或传播的数据可以识别图片的内容属性,尤其可以用于判断是否为广告图片。
Description
本发明涉及图像识别领域,具体涉及一种图片内容属性识别方法和系统。
在网络中很多类型的资源站点上,都会出现一些广告图片,这些广告图片的种类非常丰富,其包括各类商品的广告(例如,关于奶粉、衣服的广告),和实体商店的广告,以及一些其他类型的广告。
这些广告图片不但会出现在商家的站点上,也会出现在其他资源站点的页面上,例如,在允许用户上传图片的社区(论坛、图片站等),会有一些用户上传广告图片。大量广告图片的存在,往往对用户造成干扰,甚至用户进行图片搜索时,也会出现与用户需求无关的广告图片。
从图片的图像内容角度来看,不同广告图片是没有特别多的相似点的,所以基于目前的图像识别技术,难以对图片的图片内容属性进行识别,即难以识别出哪些图片为广告图片,也就无法对广告图片进行针对性的处理,用户的体验必然受到广告图片的影响。
发明内容
鉴于上述问题,提出了本发明以便提供一种克服上述问题或者至少部分地解决上述问题的一种图片内容属性识别方法和系统。
依据本发明的一个方面,提供了一种图片内容属性识别方法,其包括:计算多个同源图片簇对于特定资源站点的相对转载数;根据所述相对转载数识别对应的同源图片簇中的图片内容属性。
可选地,所述计算多个同源图片簇对于特定资源站点的相对转载数之前,还包括:对收集到的图片进行相似图片识别,将图片聚合为多个同源图片簇。
可选地,根据所述相对转载数识别对应的同源图片簇中的图片内容属性,包括:
根据所述多个同源图片簇以及对应的相对转载数训练筛选器模型;
根据训练后的筛选器模型识别目标图片簇中的图片内容属性。
依据本发明的另一个方面,提供了一种图片内容属性识别系统,其包括:相对转载数计算模块,用于计算多个同源图片簇对于特定资源站点的相对转载数;图片内容属性识别模块,用于根据所述相对转载数识别对应的同源图片簇中的图片内容属性。
根据本发明的又一个方面,提供了一种计算机程序,包括计算机可读代码,当所述计算机可读代码在计算设备上运行时,导致所述计算设备执行根据权利要求1-12中的任一个所述的图片内容属性识别方法。
根据本发明的再一个方面,提供了一种计算机可读介质,其中存储了如权利要求23所述的计算机程序。
根据本发明的图片内容属性识别方法和系统,利用了同源图片簇对于特定资源站点的相对转载数识别对应的同源图片簇中的图片内容属性,用于进行图片属性识别的数据选用相对转载数,这是因为相对转载数是能够反映图片在特定资源站点的站内站外比例的数据,而作为广告的图片的一个主要特点在于:在某一资源站点上转载的次数非常高,而在互联网范围内其他资源站点上转载的次数会相对地明显变少,因此相对转载数的大小可以用于区分别图片是否作为广告进行传播,利用相对转载数对图片的图片内容属性进行识别,可以准确地判断图片是否为广告图片。
进一步,将相似的图片聚合到同一同源图片簇中,对于一个同源图片簇而言,如果其中一张图片为广告图片,则其余图片也必然为广告图片,所以本发明的技术方案中以图片簇为单位进行图片内容属性的识别,以判断每个同源图片簇中的图片是否为广告图片。
进一步,利用了同源图片簇对于特定资源站点的相对转载数作为训练数据进行筛选器模型的训练,利用相对转载数进行的筛选器模型的训练,则得到的筛选器模型可以自行对图片的图片内容属性进行识别,准确地判断图片是否为广告图片。
上述说明仅是本发明技术方案的概述,为了能够更清楚了解本发明的技术手段,而可依照说明书的内容予以实施,并且为了让本发明的上述和其它目的、特征和优点能够更明显易懂,以下特举本发明的具体实施方式。
通过阅读下文优选实施方式的详细描述,各种其他的优点和益处对于本领域普通技术人员将变得清楚明了。附图仅用于示出优选实施方式的目的,而并不认为是对本发明的限制。而且在整个附图中,用相同的参考符号表示相同的部件。在附图中:
图1示出了根据本发明的一个实施例的图片内容属性识别方法的流程图;
图2示出了根据本发明的一个实施例的图片内容属性识别系统的框图;
图3示出了根据本发明的一个实施例的图片内容属性识别方法的流程图;
图4示出了根据本发明的一个实施例的图片内容属性识别方法的部分流程图;
图5示出了根据本发明的一个实施例的图片内容属性识别方法的流程图;
图6示出了根据本发明的一个实施例的图片内容属性识别系统的框图;
图7示出了根据本发明的一个实施例的图片内容属性识别系统的框图;
图8示出了根据本发明的一个实施例的图片内容属性识别系统的框图;
图9示出了根据本发明的一个实施例的图片内容属性识别方法的流程图;
图10示出了根据本发明的一个实施例的图片内容属性识别方法的部分流程图;
图11示出了根据本发明的一个实施例的图片内容属性识别方法的流程图;
图12示出了根据本发明的一个实施例的图片内容属性识别系统的框图;
图13示出了根据本发明的一个实施例的图片内容属性识别系统的框图;
图14示出了根据本发明的一个实施例的图片内容属性识别系统的框图;
图15示意性地示出了用于执行根据本发明的图片内容属性识别方法的计算设备的框图;以及
图16示意性地示出了用于保持或者携带实现根据本发明的图片内容属性识别方法的程序代码的存储单元。
下面结合附图和具体的实施方式对本发明作进一步的描述。
为解决上述技术问题,本发明实施例提供了一种图片内容属性识别方法。图1示出了根据本发明一个实施例的图片内容属性识别方法的处理流程图。参见图1,该方法包括:
步骤S102、计算多个同源图片簇对于特定资源站点的相对转载数;
步骤S104、根据所述相对转载数识别对应的同源图片簇中的图片内容属性。
为支持图1所示的图片内容属性识别方法,本发明实施例还提供了一种图片内容属性识别系统。图2示出了根据本发明一个实施例的图片内容属性识别系统的结构示意图。参见图2,该系统包括:
相对转载数计算模块210,用于计算多个同源图片簇对于特定资源站点的相对转载数;
图片内容属性识别模块220,与相对转载数计算模块210耦合,用于根据所述相对转载数识别对应的同源图片簇中的图片内容属性。
图1所示流程指出,根据相对转载数能够识别图片内容属性,但是如何实现这一功能有不同的手段或方式,现以不同的实施例对不同的实现方式进行详细阐述。
实施例一
如图3所示,本发明的一个实施例(实施例一)提供了一种图片内容属性识别方法,其包括:步骤310,计算多个同源图片簇对于特定资源站点的相对转载数,每个图片簇是对一组图片的聚合,例如,可以是相似度较高的一组图片,而相对转载数是一种能够反映同源图片簇的图片在特定资源站点站内站外的转载比例的数据,相对转载数的计算方式较多,本实施例中不对相对转载数的计算方式进行限制;步骤320,根据多个同源图片簇以及对应的相对转载数训练筛选器模型,通过对广告图片的研究发现,广告图片有以下特点:广告图片生产成本高,很多广告图片都是商户花费金钱、花费时间制作的,因为广告图片的生产成本高,所以商户会将一张广告图片传播很多次,但是这些广告图片基本上只有商户会进行传播,而其他的用户则基本不会传播广告图片,广告图片在传播上的这种差别最终会体现在资源站点上的转载数上:在特定的资源站点上转载的次数非常多(商户故意传播),而在互联网其他站点上的转载的次数相对少的多(其他用户并不传播),也即广告图片在特定资源站点站内站外的转载比例会比较高,所以相对转载数可以作为区分广告图片和非广告图片的一种数据,而训练筛选器模型的工具包括但不
限于开源的LIB SVM;步骤330,根据训练后的筛选器模型识别目标图片簇中的图片内容属性,即识别目标图片簇中的图片是否为广告图片,有利于对广告图片进行过滤等处理,避免广告图片对用户的体验造成影响,假设目标图片簇为对应图片搜索请求的一组图片,则根据本实施例的技术方案,可以从其中识别出广告图片并进行过滤,从而将非广告图片作为搜索结果提供给用户,从而保证用户的使用体验。
在实际应用中,在本发明提出的相对转载数之外,还同时考虑到其他的特征,例如图片的长/宽,图片的大小,图片的清晰度,图片链接是否和网页同站,或图片跳转链接是否站外等特征,在训练筛选器时会根据多个同源图片簇各自对应的相对转载数,以及图片簇中的图片的长/宽,图片的大小,图片的清晰度,图片链接是否和网页同站,图片跳转链接是否站外中的一个或多个组合,先经过筛选器去学习和训练。在目标图片簇识别时,也会对应参照上述这些其他特征中的一个或多个来进行筛选并识别是否为广告图片。
本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,步骤310可以包括:对于多个同源图片簇中的一个同源图片簇,将同源图片簇中的图片在特定资源站点上的转载数,例如在图片站A上转载了30次,与在多个资源站点上的转载数相比较,例如在10个图片站(包括图片站A)上共转载了35次,得到同源图片簇对于特定资源站点的相对转载数,多个资源站点包括特定资源站点,本实施例中提供了计算相对转载数的可行方式,且不对具体的比较方式进行限定,例如,取30/35、30/(35-30)作为相对转载数都是可以的。
如图4所示,本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,步骤310包括:步骤311,计算特定资源站点上的图片的第一平均转载数,例如假设图片站A的第一平均转载数为5;步骤312,计算多个资源站点上的图片的第二平均转载数,例如假设10个图片站(包括图片站A)的第二平均转载数为20;步骤313,取同源图片簇中的图片在特定资源站点上的转载数与第一平均转载数的第一差值,则第一差值实际上可反映同源图片簇的图片与其他图片在特定资源站点上的转载差异,差值越大则表示同源图片簇为广告图片的可能性越大,结合前述的实施例可知第一差值为30-5=25,以及取同源图片簇中的图片在多个资源站点上的转载数与第二平均转载数的第二差值,则第二差值实际上可反映同源图片簇的图片与其他图片在多个资源站点上的转载差异,差值越大表示同源图片簇为广告图片的可能性越小,结合前述的实施例可知第二差值为35-20=15,将第一差值和第二差值对比得到同源图片簇对于特定资源站点的相对转载数,本实施例中提供了另一种计算相对转载数的方式,且考虑到同源图片簇的图片与其他图片的转载差异,使得相对转载数能更好地反映图片是否为广告图片,本实施例中不对第一差值和第二差值对比方式进行限定,例如,取25/15,(25±a)/(15±b)都是可以的,a、b为常数。
本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,步骤311包括:取多个同源图片簇的图片中位于特定资源
站点上的多个图片,将多个图片的数量与多个图片对应的同源图片簇的数量进行对比,得到第一平均转载数,例如图片站A上有100张图片,该100张图片位于20个图片簇中,则第一平均转载数为100/20=5,本实施例的技术方案中提供了一种快速高效得到平均转载数的方式。
本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,步骤312包括:将多个同源图片簇的图片的数量,与多个同源图片簇的数量进行比较,得到第二平均转载数,例如10个图片站(包括图片站A)上有1000张图片,该1000张图片可聚类为50个图片簇,则第二平均转载数为1000/50=20,本实施例的技术方案中提供了一种快速高效得到平均转载数的方式。
如图5所示,本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,步骤310之前,还包括:步骤301,抓取多个资源站点上出现的图片链接(URL);步骤302,检测图片链接与同源图片簇的图片对应的链接是否相同,这反映了一张图片是否以不同的URL被转载,和/或检测图片链接对应的图片的校验信息与同源图片簇的图片的校验信息(包括但不限于MD5值)是否相同,这反映了是否存在多张相同的图片,和/或检测图片链接对应的图片与同源图片簇的图片是否存在一个或多个相同的图像特征,这反映了多张图片是否相同,或由同一张图片修改得到,本实施例中的图像特征包括但不限于轮廓特征、颜色特征、直方图特征等;步骤303,根据检测结果,确定图片链接是否为同源图片簇的图片的转载,并统计同源图片簇的图片的转载数,则本实施例中提供了一种可全面统计图片转载数的技术方案。
本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,特定资源站点为多个同源图片簇中转载每个同源图片簇的图片最多的资源站点,转载图片最多次数的站点很可能为广告图片的商户进行传播的站点,该站点对应的转载数最能够有效地反映出图片是否为广告图片。
本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,每个同源图片簇的图片对应同一源图片,且每个同源图片簇的图片与其对应的源图片具有一个或多个相同的图像特征,则在本实施例的技术方案中,每个同源图片簇的图片相同,或可以同一图片修改得到,本实施例中的图像特征包括但不限于轮廓特征、颜色特征、直方图特征等。
如图6所示,本发明的一个实施例提供了一种图片内容属性识别系统,用于支持实施例一的图片内容属性识别方法,除包括图2中的各模块还包括其他模块。具体地,其包括:相对转载数计算模块210,用于计算多个同源图片簇对于特定资源站点的相对转载数,每个图片簇是对一组图片的聚合,例如,可以是相似度较高的一组图片,而相对转载数是一种能够反映同源图片簇的图片在特定资源站点站内站外的转载比例的数据,相对转载数的计算方式较多,本实施例中不对相对转载数的计算方式进行限制;训练模块230,用于将多个同源图片簇以及对应的相对转载数输入筛选器中训练筛选器模型。通过对广告图片的研究发现,广告图片有以下特点:广告图片生产成本高,很多广告图片都
是商户花费金钱、花费时间制作的,因为广告图片的生产成本高,所以商户会将一张广告图片传播很多次,但是这些广告图片基本上只有商户会进行传播,而其他的用户则基本不会传播广告图片,广告图片在传播上的这种差别最终会体现在资源站点上的转载数上:在特定的资源站点上转载的次数非常多(商户故意传播),而在互联网其他站点上的转载的次数相对少的多(其他用户并不传播),也即广告图片在特定资源站点站内站外的转载比例会比较高,所以相对转载数可以作为区分广告图片和非广告图片的一种数据;筛选器240,适于根据训练模块得到训练后的筛选器模型,并根据模型对目标图片簇进行筛选,本实施例中使用的筛选器包括但不限于开源的LIB SVM;图片内容属性识别模块220,用于根据筛选器240对目标图片簇进行筛选,识别目标图片簇中的图片内容属性,即识别目标图片簇中的图片是否为广告图片。
另外,实际应用中所述系统进一步包括:图片格式特征模块和/或图片链接特征模块;所述图片格式特征模块,适于提取同源图片簇以及目标图片簇中包含的图片的格式特征;所述图片链接特征模块,适于提取同源图片簇以及目标图片簇中中包含的图片的链接特征;所述训练模块230进一步适于基于多个同源图片簇、对应的相对转载数以及对应的图片格式特征和/或图片链接特征,一同输入筛选器中训练筛选器模型;所述筛选器240,进一步适于根据训练后的模型,结合目标图片簇对应的相对转载数以及对应的图片格式特征和/或图片链接特征,对目标图片簇进行筛选;所述图片内容属性识别模块220,进一步用于根据所述筛选器基于目标图片簇对应的相对转载数以及对应的图片格式特征和/或图片链接特征对目标图片簇进行筛选,识别目标图片簇中的图片内容属性。
有利于对广告图片进行过滤等处理,避免广告图片对用户的体验造成影响,假设目标图片簇为对应图片搜索请求的一组图片,则根据本实施例的技术方案,可以从其中识别出广告图片并进行过滤,从而将非广告图片作为搜索结果提供给用户,从而保证用户的使用体验。
在实际应用中,在本发明提出的相对转载数之外,还考虑到其他的特征,例如图片的长/宽,图片的大小,图片的清晰度,图片链接是否和网页同站,或图片跳转链接是否站外等特征,同样先经过分类器去学习和训练。在目标图片簇识别时,也会考虑上述这些其他特征中的一个或多个来进行筛选并识别是否为广告图片。
本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,相对转载数计算模块210对于多个同源图片簇中的一个同源图片簇,将同源图片簇中的图片在特定资源站点上的转载数,例如在图片站A上转载了30次,与在多个资源站点上的转载数相比较,例如在10个图片站(包括图片站A)上共转载了35次,得到同源图片簇对于特定资源站点的相对转载数,多个资源站点包括特定资源站点,本实施例中提供了计算相对转载数的可行方式,且不对具体的比较方式进行限定,例如,取30/35、30/(35-30)作为相对转载数都是可以的。
如图7所示,本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,还包括:第一平均转载数计算模块250,用于
计算特定资源站点上的图片的第一平均转载数,例如假设图片站A的第一平均转载数为5;第二平均转载数计算模块260,用于计算多个资源站点上的图片的第二平均转载数,例如假设10个图片站(包括图片站A)的第二平均转载数为20;相对转载数计算模块210取同源图片簇中的图片在特定资源站点上的转载数与第一平均转载数的第一差值,则第一差值实际上可反映同源图片簇的图片与其他图片在特定资源站点上的转载差异,差值越大则表示同源图片簇为广告图片的可能性越大,结合前述的实施例可知第一差值为30-5=25,以及取同源图片簇中的图片在多个资源站点上的转载数与第二平均转载数的第二差值,则第二差值实际上可反映同源图片簇的图片与其他图片在多个资源站点上的转载差异,差值越大表示同源图片簇为广告图片的可能性越小,结合前述的实施例可知第二差值为35-20=15,将第一差值和第二差值对比得到同源图片簇对于特定资源站点的相对转载数,本实施例中提供了另一种计算相对转载数的方式,且考虑到同源图片簇的图片与其他图片的转载差异,使得相对转载数能更好地反映图片是否为广告图片,本实施例中不对第一差值和第二差值对比方式进行限定,例如,取25/15,(25±a)/(15±b)都是可以的,a、b为常数。
本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,第一平均转载数计算模块250取多个同源图片簇的图片中位于特定资源站点上的多个图片,将多个图片的数量与多个图片对应的同源图片簇的数量进行对比,得到第一平均转载数,例如图片站A上有100张图片,该100张图片位于20个图片簇中,则第一平均转载数为100/20=5,本实施例的技术方案中提供了一种快速高效得到平均转载数的方式。
本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,第二平均转载数计算模块260将多个同源图片簇的图片的数量,与多个同源图片簇的数量进行比较,得到第二平均转载数,例如10个图片站(包括图片站A)上有1000张图片,该1000张图片可聚类为50个图片簇,则第二平均转载数为1000/50=20,本实施例的技术方案中提供了一种快速高效得到平均转载数的方式。
如图8所示,本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,还包括:图片链接抓取模块270,用于抓取多个资源站点上出现的图片链接(URL);图片链接检测模块280,用于检测图片链接与同源图片簇的图片对应的链接是否相同,这反映了一张图片是否以不同的URL被转载,和/或检测图片链接对应的图片的校验信息与同源图片簇的图片的校验信息(包括但不限于MD5值)是否相同,这反映了是否存在多张相同的图片,和/或检测图片链接对应的图片与同源图片簇的图片是否存在一个或多个相同的图像特征,这反映了多张图片是否相同,或由同一张图片修改得到,本实施例中的图像特征包括但不限于轮廓特征、颜色特征、直方图特征等;图片转载数统计模块290,用于根据检测结果,确定图片链接是否为同源图片簇的图片的转载,并统计同源图片簇的图片的转载数,则本实施例中提供了一种可全面统计图片转载数的技术方案。
本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,特定资源站点为多个同源图片簇中转载每个同源图片簇的图片最多的资源站点,转载图片最多次数的站点很可能为广告图片的商户进行传播的站点,该站点对应的转载数最能够有效地反映出图片是否为广告图片。
本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,每个同源图片簇的图片对应同一源图片,且每个同源图片簇的图片与其对应的源图片具有一个或多个相同的图像特征,则在本实施例的技术方案中,每个同源图片簇的图片相同,或可以同一图片修改得到,本实施例中的图像特征包括但不限于轮廓特征、颜色特征、直方图特征等。
实施例二
如图9所示,本发明的一个实施例中提供了另一种图片内容属性识别方法,其包括:步骤910,对收集到的图片进行相似图片识别,将图片聚合为多个同源图片簇,本实施例中将相似的图片聚合到同一同源图片簇中,对于一个同源图片簇而言,如果其中一张图片为广告图片,则其余图片也必然为广告图片,所以本实施例中以图片簇为单位进行图片内容属性的识别,以判断每个同源图片簇中的图片是否为广告图片,基于目前的图像识别技术可以识别相似图片,本实施例不对相似图片的识别技术进行限定;步骤920,计算多个同源图片簇对于特定资源站点的相对转载数,相对转载数是一种能够反映同源图片簇的图片在特定资源站点站内站外的转载比例的数据,相对转载数的计算方式较多,本实施例中不对相对转载数的计算方式进行限制;步骤930,根据相对转载数识别对应的同源图片簇中的图片内容属性,通过对广告图片的研究发现,广告图片有以下特点:广告图片生产成本高,很多广告图片都是商户花费金钱、花费时间制作的,因为广告图片的生产成本高,所以商户会将一张广告图片传播很多次,但是这些广告图片基本上只有商户会进行传播,而其他的用户则基本不会传播广告图片,广告图片在传播上的这种差别最终会体现在资源站点上的转载数上:在特定的资源站点上转载的次数非常多(商户故意传播),而在互联网其他站点上的转载的次数相对少的多(其他用户并不传播),也即广告图片在特定资源站点站内站外的转载比例会比较高,所以相对转载数可以作为区分广告图片和非广告图片的一种数据,因此本实施例的技术方案能够识别同源图片簇中的图片是否为广告图片,有利于对广告图片进行过滤等处理,避免广告图片对用户的体验造成影响,假设同源图片簇为对应图片搜索请求的一组图片,则根据本实施例的技术方案,可以从其中识别出广告图片并进行过滤,从而将非广告图片作为搜索结果提供给用户,从而保证用户的使用体验。
在实际应用中,在本发明提出的相对转载数之外,还同时考虑到其他的特征,例如图片的长/宽,图片的大小,图片的清晰度,图片链接是否和网页同站,或图片跳转链接是否站外等特征。在同源图片簇识别时,也会考虑上述这些其他特征中的一个或多个来进行筛选并识别是否为广告图片。在实际应用的另一实施例中,也可以先经过SVM模型去学习和训练,将相对转载数以及上述其他特征中的一个或多个的组合作为参数对分类
器进行训练,并且在最后识别时也使用训练后的SVM模型及对应特征作为参数去识别。
本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,步骤920可以包括:对于多个同源图片簇中的一个同源图片簇,将同源图片簇中的图片在特定资源站点上的转载数,例如在图片站A上转载了30次,与在多个资源站点上的转载数相比较,例如在10个图片站(包括图片站A)上共转载了35次,得到同源图片簇对于特定资源站点的相对转载数,多个资源站点包括特定资源站点,本实施例中提供了计算相对转载数的可行方式,且不对具体的比较方式进行限定,例如,取30/35、30/(35-30)作为相对转载数都是可以的。
如图10所示,本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,步骤920包括:步骤921,计算特定资源站点上的图片的第一平均转载数,例如假设图片站A的第一平均转载数为5;步骤922,计算多个资源站点上的图片的第二平均转载数,例如假设10个图片站(包括图片站A)的第二平均转载数为20;步骤923,取同源图片簇中的图片在特定资源站点上的转载数与第一平均转载数的第一差值,则第一差值实际上可反映同源图片簇的图片与其他图片在特定资源站点上的转载差异,差值越大则表示同源图片簇为广告图片的可能性越大,结合前述的实施例可知第一差值为30-5=25,以及取同源图片簇中的图片在多个资源站点上的转载数与第二平均转载数的第二差值,则第二差值实际上可反映同源图片簇的图片与其他图片在多个资源站点上的转载差异,差值越大表示同源图片簇为广告图片的可能性越小,结合前述的实施例可知第二差值为35-20=15,将第一差值和第二差值对比得到同源图片簇对于特定资源站点的相对转载数,本实施例中提供了另一种计算相对转载数的方式,且考虑到同源图片簇的图片与其他图片的转载差异,使得相对转载数能更好地反映图片是否为广告图片,本实施例中不对第一差值和第二差值对比方式进行限定,例如,取25/15,(25±a)/(15±b)都是可以的,a、b为常数。
本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,步骤921包括:取多个同源图片簇的图片中位于特定资源站点上的多个图片,将多个图片的数量与多个图片对应的同源图片簇的数量进行对比,得到第一平均转载数,例如图片站A上有100张图片,该100张图片位于20个图片簇中,则第一平均转载数为100/20=5,本实施例的技术方案中提供了一种快速高效得到平均转载数的方式。
本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,步骤922包括:将多个同源图片簇的图片的数量,与多个同源图片簇的数量进行比较,得到第二平均转载数,例如10个图片站(包括图片站A)上有1000张图片,该1000张图片可聚类为50个图片簇,则第二平均转载数为1000/50=20,本实施例的技术方案中提供了一种快速高效得到平均转载数的方式。
如图11所示,本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,步骤920之前,还包括:步骤911,抓取多个
资源站点上出现的URL;步骤912,检测图片链接与同源图片簇的图片对应的链接是否相同,这反映了一张图片是否以不同的URL被转载,和/或检测图片链接对应的图片的校验信息与同源图片簇的图片的校验信息(包括但不限于MD5值)是否相同,这反映了是否存在多张相同的图片,和/或检测图片链接对应的图片与同源图片簇的图片是否存在一个或多个相同的图像特征,这反映了多张图片是否相同,或由同一张图片修改得到,本实施例中的图像特征包括但不限于轮廓特征、颜色特征、直方图特征等;步骤913,根据检测结果,确定图片链接是否为同源图片簇的图片的转载,并统计同源图片簇的图片的转载数,则本实施例中提供了一种可全面统计图片转载数的技术方案,其中步骤911与步骤910的顺序不限。
本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,特定资源站点为多个同源图片簇中转载每个同源图片簇的图片最多的资源站点,转载图片最多次数的站点很可能为广告图片的商户进行传播的站点,该站点对应的转载数最能够有效地反映出图片是否为广告图片。
本发明的另一实施例提出一种图片内容属性识别方法,与上述实施例相比,本实施例的图片内容属性识别方法,每个同源图片簇的图片对应同一源图片,且每个同源图片簇的图片与其对应的源图片具有一个或多个相同的图像特征,则在本实施例的技术方案中,每个同源图片簇的图片相同,或可以同一图片修改得到,本实施例中的图像特征包括但不限于轮廓特征、颜色特征、直方图特征等。
如图12所示,本发明的一个实施例中提供了一种图片内容属性识别系统,用于支持实施例二提供的图片内容属性识别方法,除包括图2中的各模块还包括其他模块。具体地,其包括:图片聚合模块1210,用于对收集到的图片进行相似图片识别,将图片聚合为多个同源图片簇,本实施例中将相似的图片聚合到同一同源图片簇中,对于一个同源图片簇而言,如果其中一张图片为广告图片,则其余图片也必然为广告图片,所以本实施例中以图片簇为单位进行图片内容属性的识别,以判断每个同源图片簇中的图片是否为广告图片,基于目前的图像识别技术可以识别相似图片,本实施例不对相似图片的识别技术进行限定;相对转载数计算模块210,用于计算多个同源图片簇对于特定资源站点的相对转载数,相对转载数是一种能够反映同源图片簇的图片在特定资源站点站内站外的转载比例的数据,相对转载数的计算方式较多,本实施例中不对相对转载数的计算方式进行限制;图片内容属性识别模块220,用于根据相对转载数识别对应的同源图片簇中的图片内容属性。通过对广告图片的研究发现,广告图片有以下特点:广告图片生产成本高,很多广告图片都是商户花费金钱、花费时间制作的,因为广告图片的生产成本高,所以商户会将一张广告图片传播很多次,但是这些广告图片基本上只有商户会进行传播,而其他的用户则基本不会传播广告图片,广告图片在传播上的这种差别最终会体现在资源站点上的转载数上:在特定的资源站点上转载的次数非常多(商户故意传播),而在互联网其他站点上的转载的次数相对少的多(其他用户并不传播),也即广告图片在特定资源站点站内站外的转载比例会比较高,所以相对转载数可以作为区分广告图片和非
广告图片的一种数据,因此本实施例的技术方案能够识别同源图片簇中的图片是否为广告图片,有利于对广告图片进行过滤等处理,避免广告图片对用户的体验造成影响,假设同源图片簇为对应图片搜索请求的一组图片,则根据本实施例的技术方案,可以从其中识别出广告图片并进行过滤,从而将非广告图片作为搜索结果提供给用户,从而保证用户的使用体验。
另外,实际应用中所述系统进一步包括:图片格式特征模块和/或图片链接特征模块;所述图片格式特征模块,适于提取同源图片簇中包含的图片的格式特征;所述图片链接特征模块,适于提取同源图片簇中包含的图片的链接特征;图片内容属性识别模块220,进一步适于根据同源图片簇对应的相对转载数以及对应的图片格式特征和/或图片链接特征对同源图片簇进行筛选,识别同源图片簇中的图片内容属性。
在实际应用中,在本发明提出的相对转载数之外,还考虑到其他的特征,例如图片的长/宽,图片的大小,图片的清晰度,图片链接是否和网页同站,或图片跳转链接是否站外等特征。在同源图片簇识别时,也会考虑上述这些其他特征中的一个或多个来进行筛选并识别是否为广告图片。在实际应用的另一实施例中,也可以先经过SVM模型去学习和训练,将相对转载数以及上述其他特征中的一个或多个的组合作为参数对分类器进行训练,并且在最后识别时也使用训练后的SVM模型去识别。
本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,相对转载数计算模块210对于多个同源图片簇中的一个同源图片簇,将同源图片簇中的图片在特定资源站点上的转载数,例如在图片站A上转载了30次,与在多个资源站点上的转载数相比较,例如在10个图片站(包括图片站A)上共转载了35次,得到同源图片簇对于特定资源站点的相对转载数,多个资源站点包括特定资源站点,本实施例中提供了计算相对转载数的可行方式,且不对具体的比较方式进行限定,例如,取30/35、30/(35-30)作为相对转载数都是可以的。
如图13所示,本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,还包括:第一平均转载数计算模块1240,用于计算特定资源站点上的图片的第一平均转载数,例如假设图片站A的第一平均转载数为5;第二平均转载数计算模块1250,用于计算多个资源站点上的图片的第二平均转载数,例如假设10个图片站(包括图片站A)的第二平均转载数为20;相对转载数计算模块210取同源图片簇中的图片在特定资源站点上的转载数与第一平均转载数的第一差值,则第一差值实际上可反映同源图片簇的图片与其他图片在特定资源站点上的转载差异,差值越大则表示同源图片簇为广告图片的可能性越大,结合前述的实施例可知第一差值为30-5=25,以及取同源图片簇中的图片在多个资源站点上的转载数与第二平均转载数的第二差值,则第二差值实际上可反映同源图片簇的图片与其他图片在多个资源站点上的转载差异,差值越大表示同源图片簇为广告图片的可能性越小,结合前述的实施例可知第二差值为35-20=15,将第一差值和第二差值对比得到同源图片簇对于特定资源站点的相对转载数,本实施例中提供了另一种计算相对转载数的方式,且考虑到同源图片簇的
图片与其他图片的转载差异,使得相对转载数能更好地反映图片是否为广告图片,本实施例中不对第一差值和第二差值对比方式进行限定,例如,取25/15,(25±a)/(15±b)都是可以的,a、b为常数。
本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,第一平均转载数计算模块240取多个同源图片簇的图片中位于特定资源站点上的多个图片,将多个图片的数量与多个图片对应的同源图片簇的数量进行对比,得到第一平均转载数,例如图片站A上有100张图片,该100张图片位于20个图片簇中,则第一平均转载数为100/20=5,本实施例的技术方案中提供了一种快速高效得到平均转载数的方式。
本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,第二平均转载数计算模块250将多个同源图片簇的图片的数量,与多个同源图片簇的数量进行比较,得到第二平均转载数,例如10个图片站(包括图片站A)上有1000张图片,该1000张图片可聚类为50个图片簇,则第二平均转载数为1000/50=20,本实施例的技术方案中提供了一种快速高效得到平均转载数的方式。
如图14所示,本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,还包括:图片链接抓取模块1260,用于抓取多个资源站点上出现的URL;图片链接检测模块1270,用于检测图片链接与同源图片簇的图片对应的链接是否相同,这反映了一张图片是否以不同的URL被转载,和/或检测图片链接对应的图片的校验信息与同源图片簇的图片的校验信息(包括但不限于MD5值)是否相同,这反映了是否存在多张相同的图片,和/或检测图片链接对应的图片与同源图片簇的图片是否存在一个或多个相同的图像特征,这反映了多张图片是否相同,或由同一张图片修改得到,本实施例中的图像特征包括但不限于轮廓特征、颜色特征、直方图特征等;图片转载数统计模块1280,用于根据检测结果,确定图片链接是否为同源图片簇的图片的转载,并统计同源图片簇的图片的转载数,则本实施例中提供了一种可全面统计图片转载数的技术方案。
本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,特定资源站点为多个同源图片簇中转载每个同源图片簇的图片最多的资源站点,转载图片最多次数的站点很可能为广告图片的商户进行传播的站点,该站点对应的转载数最能够有效地反映出图片是否为广告图片。
本发明的另一实施例提出一种图片内容属性识别系统,与上述实施例相比,本实施例的图片内容属性识别系统,每个同源图片簇的图片对应同一源图片,且每个同源图片簇的图片与其对应的源图片具有一个或多个相同的图像特征,则在本实施例的技术方案中,每个同源图片簇的图片相同,或可以同一图片修改得到,本实施例中的图像特征包括但不限于轮廓特征、颜色特征、直方图特征等。
在此处所提供的说明书中,说明了大量具体细节。然而,能够理解,本发明的实施例可以在没有这些具体细节的情况下实践。在一些实例中,并未详细示出公知的方法、
结构和技术,以便不模糊对本说明书的理解。
类似地,应当理解,为了精简本公开并帮助理解各个发明方面中的一个或多个,在上面对本发明的示例性实施例的描述中,本发明的各个特征有时被一起分组到单个实施例、图、或者对其的描述中。然而,并不应将该公开的方法解释成反映如下意图:即所要求保护的本发明要求比在每个权利要求中所明确记载的特征更多的特征。更确切地说,如下面的权利要求书所反映的那样,发明方面在于少于前面公开的单个实施例的所有特征。因此,遵循具体实施方式的权利要求书由此明确地并入该具体实施方式,其中每个权利要求本身都作为本发明的单独实施例。
本领域那些技术人员可以理解,可以对实施例中的设备中的模块进行自适应性地改变并且把它们设置在与该实施例不同的一个或多个设备中。可以把实施例中的模块或单元或组件组合成一个模块或单元或组件,以及此外可以把它们分成多个子模块或子单元或子组件。除了这样的特征和/或过程或者单元中的至少一些是相互排斥之外,可以采用任何组合对本说明书(包括伴随的权利要求、摘要和附图)中公开的所有特征以及如此公开的任何方法或者设备的所有过程或单元进行组合。除非另外明确陈述,本说明书(包括伴随的权利要求、摘要和附图)中公开的每个特征可以由提供相同、等同或相似目的的替代特征来代替。
此外,本领域的技术人员能够理解,尽管在此所述的一些实施例包括其它实施例中所包括的某些特征而不是其它特征,但是不同实施例的特征的组合意味着处于本发明的范围之内并且形成不同的实施例。例如,在下面的权利要求书中,所要求保护的实施例的任意之一都可以以任意的组合方式来使用。
本发明的各个部件实施例可以以硬件实现,或者以在一个或者多个处理器上运行的软件模块实现,或者以它们的组合实现。本领域的技术人员应当理解,可以在实践中使用微处理器或者数字信号处理器(DSP)来实现根据本发明实施例的图片内容属性识别系统中的一些或者全部部件的一些或者全部功能。本发明还可以实现为用于执行这里所描述的方法的一部分或者全部的设备或者装置程序(例如,计算机程序和计算机程序产品)。这样的实现本发明的程序可以存储在计算机可读介质上,或者可以具有一个或者多个信号的形式。这样的信号可以从因特网网站上下载得到,或者在载体信号上提供,或者以任何其他形式提供。
例如,图15示出了可以实现根据本发明的图片内容属性识别方法的计算设备。该计算设备传统上包括处理器1510和以存储器1520形式的计算机程序产品或者计算机可读介质。存储器1520可以是诸如闪存、EEPROM(电可擦除可编程只读存储器)、EPROM、硬盘或者ROM之类的电子存储器。存储器1520具有用于执行上述方法中的任何方法步骤的程序代码1531的存储空间1530。例如,用于程序代码的存储空间1530可以包括分别用于实现上面的方法中的各种步骤的各个程序代码1531。这些程序代码可以从一个或者多个计算机程序产品中读出或者写入到这一个或者多个计算机程序产品中。这些计算机程序产品包括诸如硬盘,紧致盘(CD)、存储卡或者软盘之类的程序代码载体。这样
的计算机程序产品通常为如参考图16所述的便携式或者固定存储单元。该存储单元可以具有与图15的计算设备中的存储器1520类似布置的存储段、存储空间等。程序代码可以例如以适当形式进行压缩。通常,存储单元包括计算机可读代码1531’,即可以由例如诸如1510之类的处理器读取的代码,这些代码当由计算设备运行时,导致该计算设备执行上面所描述的方法中的各个步骤。
本文中所称的“一个实施例”、“实施例”或者“一个或者多个实施例”意味着,结合实施例描述的特定特征、结构或者特性包括在本发明的至少一个实施例中。此外,请注意,这里“在一个实施例中”的词语例子不一定全指同一个实施例。
应该注意的是上述实施例对本发明进行说明而不是对本发明进行限制,并且本领域技术人员在不脱离所附权利要求的范围的情况下可设计出替换实施例。在权利要求中,不应将位于括号之间的任何参考符号构造成对权利要求的限制。单词“包含”不排除存在未列在权利要求中的元件或步骤。位于元件之前的单词“一”或“一个”不排除存在多个这样的元件。本发明可以借助于包括有若干不同元件的硬件以及借助于适当编程的计算机来实现。在列举了若干装置的单元权利要求中,这些装置中的若干个可以是通过同一个硬件项来具体体现。单词第一、第二、以及第三等的使用不表示任何顺序。可将这些单词解释为名称。
此外,还应当注意,本说明书中使用的语言主要是为了可读性和教导的目的而选择的,而不是为了解释或者限定本发明的主题而选择的。因此,在不偏离所附权利要求书的范围和精神的情况下,对于本技术领域的普通技术人员来说许多修改和变更都是显而易见的。对于本发明的范围,对本发明所做的公开是说明性的,而非限制性的,本发明的范围由所附权利要求书限定。
Claims (25)
- 一种图片内容属性识别方法,其包括:计算多个同源图片簇对于特定资源站点的相对转载数;根据所述相对转载数识别对应的同源图片簇中的图片内容属性。
- 根据权利要求1所述的图片内容属性识别方法,其中,所述计算多个同源图片簇对于特定资源站点的相对转载数之前,还包括:对收集到的图片进行相似图片识别,将图片聚合为多个同源图片簇。
- 根据权利要求1所述的图片内容属性识别方法,其中,根据所述相对转载数识别对应的同源图片簇中的图片内容属性,包括:根据所述多个同源图片簇以及对应的相对转载数训练筛选器模型;根据训练后的筛选器模型识别目标图片簇中的图片内容属性。
- 根据权利要求1所述的图片内容属性识别方法,其中,所述计算多个同源图片簇对于特定资源站点的相对转载数的步骤包括:对于所述多个同源图片簇中的一个同源图片簇,将所述同源图片簇中的图片在特定资源站点上的转载数,与在多个资源站点上的转载数相比较,得到所述同源图片簇对于所述特定资源站点的相对转载数,所述多个资源站点包括所述特定资源站点。
- 根据权利要求4所述的图片内容属性识别方法,其中,所述将所述同源图片簇中的图片在所述特定资源站点上的转载数,与在多个资源站点上的转载数相比较的步骤包括:计算所述特定资源站点上的图片的第一平均转载数;计算所述多个资源站点上的图片的第二平均转载数;取所述同源图片簇中的图片在所述特定资源站点上的转载数与所述第一平均转载数的第一差值,以及取所述同源图片簇中的图片在所述多个资源站点上的转载数与所述第二平均转载数的第二差值,将所述第一差值和所述第二差值对比得到所述同源图片簇对于所述特定资源站点的相对转载数。
- 根据权利要求5所述的图片内容属性识别方法,其中,所述计算所述特定资源站点上的图片的第一平均转载数的步骤包括:取所述多个同源图片簇的图片中位于所述特定资源站点上的多个图片,将所述多个图片的数量与所述多个图片对应的同源图片簇的数量进行对比,得到所述第一平均转载数。
- 根据权利要求5所述的图片内容属性识别方法,其中,所述计算所述多个资源站点上的图片的第二平均转载数的步骤包括:将所述多个同源图片簇的图片的数量,与所述多个同源图片簇的数量进行比较,得到所述第二平均转载数。
- 根据权利要求4所述的图片内容属性识别方法,其中,在所述将所述同源图片簇 中的图片在特定资源站点上的转载数,与在多个资源站点上的转载数相比较的步骤之前,还包括:抓取所述多个资源站点上出现的图片链接;检测所述图片链接与所述同源图片簇的图片对应的链接是否相同,和/或检测所述图片链接对应的图片的校验信息与所述同源图片簇的图片的校验信息是否相同,和/或检测所述图片链接对应的图片与所述同源图片簇的图片是否存在一个或多个相同的图像特征;根据检测结果,确定所述图片链接是否为所述同源图片簇的图片的转载,并统计所述同源图片簇的图片的转载数。
- 根据权利要求4所述的图片内容属性识别方法,其中,所述特定资源站点为所述多个同源图片簇中转载每个同源图片簇的图片最多的资源站点。
- 根据权利要求1至9中任一项所述的图片内容属性识别方法,其中,每个同源图片簇的图片对应同一源图片,且每个同源图片簇的图片与其对应的源图片具有一个或多个相同的图像特征。
- 根据权利要求1至9任一项所述的方法,其中,所述方法进一步包括:提取所述同源图片簇中包含的图片的格式特征和/或图片的链接特征,根据所述多个同源图片簇、对应的相对转载数,以及对应包含的图片的格式特征训练筛选器模型;根据训练后的筛选器模型,基于所述相对转载数以及目标图片簇中包含的图片的格式特征和/或图片的链接特征,来识别目标图片簇中的图片内容属性。
- 根据权利要求1至9任一项所述的方法,其中,所述图片的格式特征包括但不限于以下中的一种或几种组合:图片的长/宽,图片的大小,图片的清晰度。
- 根据权利要求1至9任一项所述的方法,其中,所述图片的链接特征包括但不限于以下中的一种或几种组合:图片链接是否和网页同站,图片跳转链接是否站外。
- 一种图片内容属性识别系统,其包括:相对转载数计算模块,用于计算多个同源图片簇对于特定资源站点的相对转载数;图片内容属性识别模块,用于根据所述相对转载数识别对应的同源图片簇中的图片内容属性。
- 根据权利要求14所述的图片内容属性识别系统,其中,还包括:图片聚合模块,用于对收集到的图片进行相似图片识别,将图片聚合为多个同源图片簇。
- 根据权利要求14所述的图片内容属性识别系统,其中,还包括:训练模块,用于将所述多个同源图片簇以及对应的相对转载数输入筛选器中训练筛选器模型;筛选器,适于根据所述训练模块得到训练后的筛选器模型,并根据所述模型对目标图片簇进行筛选;所述图片内容属性识别模块,还用于根据所述筛选器对目标图片簇进行筛选,识别目标图片簇中的图片内容属性。
- 根据权利要求14所述的图片内容属性识别系统,其中,所述相对转载数计算模块对于所述多个同源图片簇中的一个同源图片簇,将所述同源图片簇中的图片在特定资源站点上的转载数,与在多个资源站点上的转载数相比较,得到所述同源图片簇对于所述特定资源站点的相对转载数,所述多个资源站点包括所述特定资源站点。
- 根据权利要求17所述的图片内容属性识别系统,其中,还包括:第一平均转载数计算模块,用于计算特定资源站点上的图片的第一平均转载数;第二平均转载数计算模块,用于计算多个资源站点上的图片的第二平均转载数;相对转载数计算模块取同源图片簇中的图片在特定资源站点上的转载数与第一平均转载数的第一差值,以及取同源图片簇中的图片在多个资源站点上的转载数与第二平均转载数的第二差值,将第一差值和第二差值对比得到同源图片簇对于特定资源站点的相对转载数。
- 根据权利要求18所述的图片内容属性识别系统,其中,第一平均转载数计算模块取多个同源图片簇的图片中位于特定资源站点上的多个图片,将多个图片的数量与多个图片对应的同源图片簇的数量进行对比,得到第一平均转载数。
- 根据权利要求18所述的图片内容属性识别系统,其中,第二平均转载数计算模块将多个同源图片簇的图片的数量,与多个同源图片簇的数量进行比较,得到第二平均转载数。
- 根据权利要求14至20任一项所述的图片内容属性识别系统,其中,还包括:图片链接抓取模块,用于抓取多个资源站点上出现的图片链接;图片链接检测模块,用于检测图片链接与同源图片簇的图片对应的链接是否相同,和/或检测图片链接对应的图片的校验信息与同源图片簇的图片的校验信息是否相同,和/或检测图片链接对应的图片与同源图片簇的图片是否存在一个或多个相同的图像特征;图片转载数统计模块,用于根据检测结果,确定图片链接是否为同源图片簇的图片的转载,并统计同源图片簇的图片的转载数。
- 根据权利要求14至20任一项所述的图片内容属性识别系统,其中,特定资源站点为多个同源图片簇中转载每个同源图片簇的图片最多的资源站点。
- 根据权利要求14至20任一项所述的图片内容属性识别系统,其中,每个同源图片簇的图片对应同一源图片,且每个同源图片簇的图片与其对应的源图片具有一个或多个相同的图像特征。
- 一种计算机程序,包括计算机可读代码,当所述计算机可读代码在计算设备上运行时,导致所述计算设备执行根据权利要求1-13中的任一个所述的图片内容属性识别方法。
- 一种计算机可读介质,其中存储了如权利要求24所述的计算机程序。
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310632631.0 | 2013-12-02 | ||
| CN201310632631.0A CN103617261B (zh) | 2013-12-02 | 2013-12-02 | 图片内容属性识别方法和系统 |
| CN201310632676.8A CN103617262B (zh) | 2013-12-02 | 2013-12-02 | 图片内容属性识别方法和系统 |
| CN201310632676.8 | 2013-12-02 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2015081748A1 true WO2015081748A1 (zh) | 2015-06-11 |
Family
ID=53272845
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2014/087109 Ceased WO2015081748A1 (zh) | 2013-12-02 | 2014-09-22 | 图片内容属性识别方法和系统 |
Country Status (1)
| Country | Link |
|---|---|
| WO (1) | WO2015081748A1 (zh) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113449213A (zh) * | 2020-03-27 | 2021-09-28 | 北京沃东天骏信息技术有限公司 | 页面跳转的方法和装置 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101071433A (zh) * | 2007-05-10 | 2007-11-14 | 腾讯科技(深圳)有限公司 | 一种图片下载系统及方法 |
| CN102419777A (zh) * | 2012-01-10 | 2012-04-18 | 凤凰在线(北京)信息技术有限公司 | 一种互联网图片广告过滤系统及其过滤方法 |
| CN103617261A (zh) * | 2013-12-02 | 2014-03-05 | 北京奇虎科技有限公司 | 图片内容属性识别方法和系统 |
| CN103617262A (zh) * | 2013-12-02 | 2014-03-05 | 北京奇虎科技有限公司 | 图片内容属性识别方法和系统 |
-
2014
- 2014-09-22 WO PCT/CN2014/087109 patent/WO2015081748A1/zh not_active Ceased
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101071433A (zh) * | 2007-05-10 | 2007-11-14 | 腾讯科技(深圳)有限公司 | 一种图片下载系统及方法 |
| CN102419777A (zh) * | 2012-01-10 | 2012-04-18 | 凤凰在线(北京)信息技术有限公司 | 一种互联网图片广告过滤系统及其过滤方法 |
| CN103617261A (zh) * | 2013-12-02 | 2014-03-05 | 北京奇虎科技有限公司 | 图片内容属性识别方法和系统 |
| CN103617262A (zh) * | 2013-12-02 | 2014-03-05 | 北京奇虎科技有限公司 | 图片内容属性识别方法和系统 |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113449213A (zh) * | 2020-03-27 | 2021-09-28 | 北京沃东天骏信息技术有限公司 | 页面跳转的方法和装置 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| TWI753044B (zh) | 影片資料的推薦方法和裝置 | |
| WO2017000109A1 (zh) | 搜索方法、搜索装置、用户设备和计算机程序产品 | |
| CN108520196A (zh) | 奢侈品辨别方法、电子装置及存储介质 | |
| US11853368B2 (en) | Method and system for identifying and displaying an object | |
| JP6661036B2 (ja) | 閲覧者行動に基づいてメディアコンテンツをベンチマークする方法 | |
| CN110866489B (zh) | 图像识别方法、装置、设备及存储介质 | |
| CN103617192B (zh) | 一种数据对象的聚类方法和装置 | |
| JP2017514211A5 (zh) | ||
| CN113254696A (zh) | 一种封面图像获取方法及装置 | |
| CN103617262B (zh) | 图片内容属性识别方法和系统 | |
| CN106445977A (zh) | 图片推送方法及装置 | |
| CN107784507A (zh) | 疑似侵权商品预警方法及装置、计算机存储介质及设备 | |
| CN110363206B (zh) | 数据对象的聚类、数据处理及数据识别方法 | |
| Ilakkuvan et al. | Cameras for public health surveillance: a methods protocol for crowdsourced annotation of point-of-sale photographs | |
| CN103617261B (zh) | 图片内容属性识别方法和系统 | |
| CN109102324B (zh) | 模型训练方法、基于模型的红包物料铺设预测方法及装置 | |
| CN106649738A (zh) | 在搜索结果页中聚合人物类资讯信息的方法及装置 | |
| CN105653548A (zh) | 一种电子文档页面类型识别方法和系统 | |
| English et al. | Image Processing for Public Health Surveillance of Tobacco Point-of-Sale advertising: machine learning–based methodology | |
| CN103870671B (zh) | 一种从Cookie中提取用户样本的方法及装置 | |
| US20160027050A1 (en) | Method of providing advertisement service using cloud album | |
| CN109660871A (zh) | 一种弹幕角色信息确定方法、装置及设备 | |
| CN111414074A (zh) | 屏幕浏览数据处理方法、装置、介质及电子设备 | |
| CN102929948B (zh) | 列表页识别系统及方法 | |
| CN115115901B (zh) | 获取跨域学习模型的方法及装置 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 14868492 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 14868492 Country of ref document: EP Kind code of ref document: A1 |