Classification of images as advertisement images or non-advertisement images of web pages

• US 7,840,502 B2
• Filed: 06/13/2007
• Issued: 11/23/2010
• Est. Priority Date: 06/13/2007
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Li, Mingjing, Li, Dongfang, Wang, Bin, Li, Zhiwei
• Primary Examiner(s): Holmes; Michael B.
• Assistant Examiner(s): CHANG, LI WU
• Application Number:
  US11/762,553
• Publication Number:
  US 20080313031A1
• Time in Patent Office: 1,259 Days
• Field of Search: 706/12
• US Class Current: 706/12
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0277 Online advertisement

Abstract

An advertisement image classification system trains a binary classifier to classify images as advertisement images or non-advertisement images and then uses the binary classifier to classify images of web pages as advertisement images or non-advertisement images. During a training phase, the classification system generates training data of feature vectors representing the images and labels indicating whether an image is an advertisement image or a non-advertisement image. The classification system trains a binary classifier to classify images using training data. During a classification phase, the classification system inputs a web page with an image and generates a feature vector for the image. The classification system then applies the trained binary classifier to the feature vector to generate a score indicating whether the image is an advertisement image or a non-advertisement image.

Claims

1. A method in a computing device for identifying advertisement images of web pages, the method comprising:

providing training images of web pages, each provided training image being referenced by a web page;

labeling the images as advertisement images or non-advertisement images;

generating a feature vector for each of the training images, the feature vector including visual layout features derived from the web page of the image and content features derived from content of the image, the content features being selected from the group consisting of number of different colors in the content of the image, an indication of whether the image has high contrast, and an indication of whether the image is a photograph;

training a binary classifier using the feature vectors and labels of the images; and

classifying an image as an advertisement image or non-advertisement image by generating a feature vector for the image and applying the trained binary classifier to the generated feature vector of the image.

2. The method of claim 1 including learning relevant features for classifying images as advertisement images and non-advertisement images.

3. The method of claim 2 wherein the learning includes applying an adaptive boosting technique to learn weights of weak classifiers, each weak classifier corresponding to a feature, and selecting as relevant features those features whose weights indicate they are effective at distinguishing advertisement images from non-advertisement images.

4. The method of claim 1 wherein the features of an image include visual layout features and content features.

5. The method of claim 4 wherein the visual layout features include features selected from the group consisting of coordinates of top, right, and left sides of the image, coordinates of center of the image, width of the image, number of images of the web page with same right coordinate as the image, number of images of the web page with same top coordinate as the image, number of images of the web page with same right coordinate and same width as the image, and number of images of the web page with same top coordinate and same height as the image.

6. The method of claim 4 wherein the content features include features selected from the group consisting of size of the image and percentage of gray of the image.

7. The method of claim 4 wherein the features of an image further include features derived from hyperlinks.

8. The method of claim 1 wherein the binary classifier is trained using a support vector machine.

9. The method of claim 8 including learning relevant features for classifying images as advertisement images and non-advertisement images by applying an adaptive boosting technique to learn weights for weak classifiers, each weak classifier corresponding to a feature, and selecting as relevant features those features whose weights indicate they are effective at distinguishing advertisement images from non-advertisement images and wherein the features of an image include visual layout features and content features.

10. The method of claim 1 wherein features of an image to include in the feature vectors are learned by training a classifier using candidate features and selecting as relevant features those features that are most heavily weighted by the classifier.

11. A computer-readable medium for generating a binary classifier for classifying images of web pages as advertisement images or non-advertisement images, by a method comprising:

providing training web pages;

identifying the images of the training web pages;

receiving labels for the images indicating whether an image is an advertisement image or non-advertisement image;

generating a feature vector for each of the identified images, the feature vector including visual layout features of the image on the web page and content features derived from content of the image by accessing the content of the image, the content features being selected from the group consisting of aspect ratio of the image, image format, an indication of whether the image is a photograph or a graphic, size of the image, number of different colors of the image, percentage of gray area of the image, and an indication of whether the image has high contrast; and

training a binary classifier using the feature vectors and labels of the images wherein the training identifies weights of the features for use in classifying images.

12. The computer-readable medium of claim 11 including classifying an image as an advertisement image or non-advertisement image by generating a feature vector for the image and applying the trained binary classifier to the generated feature vector of the image.

13. The computer-readable medium of claim 11 including learning relevant features for classifying images as advertisement images and non-advertisement images.

14. The computer-readable medium of claim 13 wherein the learning includes applying an adaptive boosting technique to learn weights of feature-specific weak classifiers and selecting as relevant features those features of weak classifiers whose weights indicate they are effective at distinguishing advertisement images from non-advertisement images.

15. The computer-readable medium of claim 11 wherein the features of an image include content features.

16. The computer-readable medium of claim 11 wherein the binary classifier is trained using a support vector machine.

17. A computing device for identifying features of images of web pages for use in classifying images as advertisement images or non-advertisement images, comprising:

a component that retrieves images of web pages and stores in a training data store feature vectors for the retrieved images of the web pages, the retrieved images being labeled as advertisement images or non-advertisement images, the feature vectors including candidate features that include visual layout features derived from the web page of the image and content features derived from content of the retrieved images by accessing the content of the retrieved images, the content features being selected from the group consisting of aspect ratio of the image, image format, an indication of whether the image is a photograph or a graphic, size of the image, number of different colors of the image, percentage of gray area of the image, and an indication of whether the image has high contrast;

a component that trains a binary classifier using the feature vectors with candidate features and labels of the training data store; and

a component that selects as features for use in classifying images those candidate features whose weights indicate they are effective at distinguishing advertisement images from non-advertisement images.

18. The computing device of claim 17 including a component that trains a binary classifier to classify images as advertisement images and non-advertisement images using the selected features.

19. The computing device of claim 18 wherein the classifier trained using the candidate features is trained using an adaptive boosting technique and the binary classifier trained using the selected features is trained using a support vector machine technique.

20. The computing device of claim 19 including a component that classifies an image as an advertisement image or non-advertisement image by generating a feature vector of the selected features for the image and applying the trained binary classifier to the generated feature vector of the image.