MULTIMODAL CLASSIFICATION OF ADULT CONTENT

• US 20090034851A1
• Filed: 08/03/2007
• Published: 02/05/2009
• Est. Priority Date: 08/03/2007
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Qian, Richard, Fan, Xiadong
• Granted Patent: US 8,000,504 B2
• Time in Patent Office: Days
• US Class Current: 382/230
• CPC Class Codes: G06F 16/435 Filtering based on additional data, e.g. user or group profiles
• CPC Class Codes: G06F 16/583 using metadata automatically derived from the content
• CPC Class Codes: G06F 18/24323 Tree-organised classifiers
• CPC Class Codes: G06F 18/256 of results relating to different input data, e.g. multimodal recognition
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0603 Catalogue ordering

Abstract

Systems and methods for classifying content as adult content and, if desired, blocking content so classified from presentation to a user are provided. Received content is analyzed using a sequential series of classification techniques, each successive technique being implemented only if the previous technique did not result in classification of the content as adult content. In this way, adult content may be identified across a variety of different media types (e.g., text, images, video, etc.) and yet processing power may be reserved if one or more techniques requiring less power is sufficient to determine that the received content is, in fact, adult content. Content classification may be performed in-band (that is, in substantially real-time such that content may be identified and/or blocked at the time results of a user query are returned) or out-of-band (that is, prospectively as new content is received but not in association with a user query).

Claims

1. One or more computer storage media having computer-executable instructions embodied thereon for performing a method for classifying adult content, the method comprising:

receiving content having a Uniform Resource Locator (URL), webpage text, and at least one image associated therewith;

analyzing one or more of the URL, the webpage text, and the at least one image to determine a likelihood that the received content is adult content, wherein the webpage text is analyzed only if the URL is not indicative of adult content, and wherein the at least one image is analyzed only if the webpage text is not indicative of adult content; and

classifying the received content as adult or not adult based upon the determined likelihood.

2. The one or more computer storage media of claim 1, wherein the method further comprises:

receiving a search request;

determining that the received content satisfies the search request; and

presenting the received content in response to the search request only if the received content is classified as not adult.

3. The one or more computer storage media of claim 1, wherein analyzing one or more of the URL, the webpage text, and the at least one image to determine a likelihood that the received content is adult content comprises:

analyzing the URL to determine a first likelihood that the received content is adult content; and

determining whether the first likelihood exceeds a first threshold,

wherein if the first likelihood exceeds the first threshold, the determined likelihood and the first likelihood are equivalent,

and wherein if the first likelihood does not exceed the first threshold, the method further comprises analyzing the webpage text.

4. The one or more computer storage media of claim 3, wherein analyzing the webpage text comprises:

analyzing the webpage text to determine a second likelihood that the received content is adult content; and

determining whether the second likelihood exceeds a second threshold,

wherein if the second likelihood exceeds the second threshold, the determined likelihood and the second likelihood are equivalent,

and wherein if the second likelihood does not exceed the second threshold, the method further comprises analyzing the at least one image.

5. The one or more computer storage media of claim 4, wherein analyzing the at least one image comprises:

analyzing the at least one image to determine a third likelihood that the received content is adult content; and

determining whether a third likelihood exceeds a third threshold,

wherein if the third likelihood exceeds the third threshold, the determined likelihood and the third likelihood are equivalent,

and wherein if the third likelihood does not exceed the third threshold, determining the likelihood that the received content is adult content comprises collectively analyzing the first likelihood, the second likelihood, and the third likelihood.

6. The one or more computer storage media of claim 3, wherein analyzing the URL to determine the first likelihood that the received content is adult content comprises:

breaking the URL into one or more words;

normalizing the one or more words utilizing at least one of stemming and equivalence classifying;

comparing the normalized one or more words to a list of known adult words; and

determining the first likelihood based upon at least one of a number and a frequency of the normalized words included on the list.

7. The one or more computer storage media of claim 6, wherein breaking the URL into one or more words comprises:

identifying at least one bi-gram by combining consecutive words of the one or more words; and

comparing the at least one identified bi-gram to a list of known adult bi-grams,

wherein determining the first likelihood comprises determining the first likelihood based upon at least one of a number and a frequency of the normalized words included on the list of known adult words and a number and a frequency of identified bi-grams included on the list of known adult bi-grams.

8. The one or more computer storage media of claim 4, wherein analyzing the webpage text to determine a second likelihood that the received content is adult content comprises:

breaking the webpage text into one or more words, the webpage text including at least one of webpage content and a textual description of an image and/or video content;

normalizing the one or more words utilizing at least one of stemming and equivalence classifying;

comparing the normalized one or more words to a list of known adult words; and

determining the first likelihood based upon at least one of a number and a frequency of the normalized words included on the list.

9. The one or more computer storage media of claim 8, wherein breaking the webpage text into one or more words comprises:

identifying at least one bi-gram by combining consecutive words of the one or more words; and

comparing the at least one identified bi-gram to a list of known adult bi-grams,

wherein determining the first likelihood comprises determining the first likelihood based upon at least one of a number and a frequency of the normalized words included on the list of known adult words and a number and a frequency of identified bi-grams included on the list of known adult bi-grams.

10. The one or more computer storage media of claim 5, wherein analyzing the at least one image comprises:

analyzing the at least one image to determine if the at least one image includes one or more face regions;

analyzing the at least one image to determine if the at least one image includes one or more regions of predominantly skin-colored pixels;

analyzing the at least one image to determine if the at least one image includes one or more highly textured regions; and

determining the third likelihood based upon whether the at least one image includes one or more face regions, one or more regions of predominantly skin-colored pixels, and one or more highly textured regions.

11. The one or more computer storage media of claim 10,

wherein if it is determined that the at least one image includes one or more regions of predominantly skin-colored pixels, the method further comprises analyzing a shape of the one or more regions of predominantly skin-colored pixels to determine if at least one of the one or more regions of predominantly skin-colored pixels has a human-body shape,

and wherein determining the third likelihood comprises determining the third likelihood based upon whether the at least one image includes one or more face regions, one or more highly textured regions, and one or more human-body shaped regions of predominantly skin-colored pixels.

12. A method for classifying adult content, comprising:

receiving content having at least one image associated therewith;

analyzing the at least one image to determine if the at least one image includes one or more face regions;

analyzing the at least one image to determine if the at least one image includes one or more regions of predominantly skin-colored pixels;

analyzing the at least one image to determine if the at least one image includes one or more highly textured regions;

determining a likelihood that the at least one image is an adult image based upon whether the at least one image includes one or more face regions, one or more regions of predominantly skin-colored pixels, and one or more highly textured regions; and

classifying the received content as adult or not adult based upon the determined likelihood with respect to the at least one image.

13. The method of claim 12, further comprising:

receiving a search request;

determining that the received content satisfies the search request; and

presenting the received content in response to the search request only if the received content is classified as not adult.

14. The method of claim 12,

wherein if it is determined that the at least one image includes one or more regions of predominantly skin-colored pixels, the method further comprises analyzing a shape of the one or more regions of predominantly skin-colored pixels to determine if at least one of the one or more regions of predominantly skin-colored pixels has a human-body shape,

and wherein determining a likelihood that the at least one image is an adult image comprises determining a likelihood that the at least one image is an adult image based upon whether the at least one image includes one or more face regions, one or more highly textured regions, and one or more human-body shaped regions of predominantly skin-colored pixels.

15. The method of claim 12, wherein if it is determined that the at least one image includes one or more regions of predominantly skin-colored pixels and it is determined that the at least one image includes one or more face regions, the method further comprises:

comparing the one or more regions of predominantly skin-colored pixels and the one or more face regions to determine coincidence there between,

wherein if it is determine that at least one of the one or more face regions is coincident with at least one of the one or regions of predominantly skin-colored pixels, the coincident region is removed with respect to determining the likelihood that the at least one image is an adult image.

16. The method of claim 12, wherein if it is determined that the at least one image includes one or more regions of predominantly skin-colored pixels and it is determined that the at least one image includes one or more highly textured regions, the method further comprises:

comparing the one or more regions of predominantly skin-colored pixels and the one or more highly textured regions to determine coincidence there between,

wherein if it is determine that at least one of the one or more highly textured regions is coincident with at least one of the one or regions of predominantly skin-colored pixels, the coincident region is removed with respect to determining the likelihood that the at least one image is an adult image.

17. The method of claim 12, wherein receiving content having at least one image associated therewith comprises receiving content having a Uniform Resource Locator (URL), webpage text, and at least one image associated therewith, and wherein the method further comprises:

analyzing one or more of the URL and the webpage text to determine a likelihood that the received content is adult content, wherein the webpage text is analyzed only if the URL is not indicative of adult content,

and wherein classifying the received content as adult or not adult comprises classifying the received content as adult or not adult based upon the determined likelihood with respect to the URL and/or webpage text and the determined likelihood with respect to the at least one image.

18. The method of claim 12, further comprising:

receiving a search request;

determining that the received content satisfies the search request; and

presenting the received content in response to the search request only if the received content is classified as not adult.

20. The computer system of claim 16, wherein if the received web page is classified as adult, a Uniform Resource Locator (URL) associated with the web page is included on a list of known adult URLs.

19. A computer system embodied on at least one computer storage media having computer-executable instructions embodied thereon for performing a method for classifying web pages as containing adult content, the system comprising:

a receiving component configured for receiving a web page having a plurality of images associated therewith;

at least one analyzing component configured for analyzing each of the plurality of images to determine a likelihood that the analyzed image is an adult image;

at least one comparing component configured for comparing the determined likelihood with respect to each analyzed image to a threshold; and

at least one classifying component configured for classifying the received web page as adult or not adult based upon at least one of a number and a frequency of determined likelihoods which exceed the threshold.