System and method for image classification

• US 7,991,715 B2
• Filed: 06/27/2008
• Issued: 08/02/2011
• Est. Priority Date: 12/27/2007
• Status: Expired due to Fees

• Current Assignee: Arbor Labs, Inc.
• Original Assignee: Arbor Labs, Inc.
• Inventors: Schiff, Jeremy, Fiedler, Justin, Antonelli, Dominic, Liebowitz, Heston, Warren, Neil, Burgstone, Jonathan, Shirazi, Sharam
• Primary Examiner(s): STARKS, WILBERT L
• Application Number:
  US12/163,976
• Publication Number:
  US 20090172030A1
• Time in Patent Office: 1,131 Days
• Field of Search: 706/20, 706/45
• US Class Current: 706/20
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0244 Optimization
• CPC Class Codes: G06Q 30/0273 Determination of fees for advertising

Abstract

Systems and methods for image classification are disclosed. In one aspect, embodiments of the present disclosure include a method, which may be implemented on a system, of selecting a predetermined number of training images that are representative of images associable with a particular topic category. One embodiment can include, extracting training image features from the training images, generating a set of descriptors characteristic of images associable with the particular topic category, and generating the particular set of predetermined models that correspond to the particular topic category based on the set of descriptors.

Claims

1. A method of image classification, comprising:

selecting a predetermined number of training images that are representative of images associable with a particular topic category;

extracting training image features from the training images;

generating a set of descriptors characteristic of images associable with the particular topic category; and

generating the particular set of predetermined models that correspond to the particular topic category based on the set of descriptors.

2. The method of claim 1, further comprising, computing statistical parameters for at least a portion of the training image features of the set of image features.

3. The method of claim 1, further comprising, generating pluralities of sets of predetermined models for a plurality of topic categories.

4. The method of claim 1, further comprising,

receiving an image;

identifying a set of image features of the image;

computing statistical parameters for a portion of the set of image features;

using the statistical parameters as variables in the pluralities of sets of predetermined models; and

determining one or more topic categories of the plurality of categories that the image is associable with.

5. The method of claim 4, further comprising, assigning weighting values to each of the one or more topic categories that indicate likelihood that the image belongs to each of the one or more topic categories.

6. The method of claim 1, further comprising, selecting a second set of training images that are variants of the predetermined number of training images associable with the particular topic category.

7. The method of claim 6, wherein the variants comprise one or more of, a scaled version, a thumbnail version, and a rotated version.

8. A system for image classification, comprising:

means for, selecting a predetermined number of training images that are representative of images associable with a particular topic category;

means for, extracting training image features from the training images;

means for, generating a set of descriptors characteristic of images associable with the particular topic category; and

means for, generating the particular set of predetermined models that correspond to the particular topic category based on the set of descriptors.

9. A method of classification of an image as associated with one or more categories of a plurality of categories, comprising:

performing machine learning on a learning set of images;

generating a model based on the machine learning of the learning set of images;

determining an accuracy metric of the model using a verification set of images as one or more parameters in the model; and

generating a set of probability values;

wherein each of the probability value of the set of probability values is generated for a pair of categories based on the accuracy metric.

10. The method of claim 9, further comprising, selecting the learning set of images and the verification set of images.

11. The method of claim 10, wherein, the learning set of images and the verification set of images are selected via a k-fold cross validation process.

12. The method of claim 11, further comprising, generating a tree structure from the set of probability values.

13. The method of claim 12, further comprising, merging n categories of the plurality of categories to create a new category.

14. The method of claim 13, wherein the n categories are associated with the n-lowest probability values of the set of probability values.

15. The method of claim 13, wherein n=2.

16. The method of claim 15, further comprising,

generating a revised model based on the machine learning including the new category;

and

determining an updated accuracy metric of the revised model including the new category.

17. The method of claim 13, further comprising, performing the machine learning including the new category.

18. The method of claim 12, further comprising, selecting at least one category of the plurality of categories associable with the image based on the tree structure.

19. A system for image classification into a plurality of categories, comprising:

means for, performing machine learning on a learning set of images;

means for, generating a model based on the machine learning of the learning set of images;

means for, determining an accuracy metric of the model using a verification set of images as one or more parameters in the model; and

means for, generating a set of probability values;

wherein each of the probability value of the set of probability values is generated for a pair of categories based on the accuracy metric.

20. The system of claim 19, further comprising, means for, selecting the learning set of images and the verification set of images.

21. The system of claim 20, wherein, the learning set of images and the verification set of images are selected via a k-fold cross validation process.

22. The system of claim 19, further comprising, means for, generating a tree structure from the set of probability values.

23. The system of claim 19, further comprising, means for, merging n categories of the plurality of categories to create a new category.

24. The system of claim 23, further comprising, means for, performing the machine learning including the new category.

25. The system of claim 24, further comprising,

means for, generating a revised model based on the machine learning including the new category;

and

means for, determining an updated accuracy metric of the revised model including the new category.

26. The system of claim 19, wherein the n categories are associated with the n-lowest probability values of the set of probability values.

27. The system of claim 19, further comprising, means for, selecting at least one category of the plurality of categories associable with the image based on the tree structure.