Near-duplicate image detection

• US 8,792,728 B2
• Filed: 09/27/2010
• Issued: 07/29/2014
• Est. Priority Date: 09/27/2010
• Status: Active Grant

• Current Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Original Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Inventors: Tang, Feng, Gao, Yuli, Lim, Suk Hwan
• Primary Examiner(s): Koziol, Stephen R
• Assistant Examiner(s): CONNER, SEAN M
• Application Number:
  US12/891,234
• Publication Number:
  US 20120076423A1
• Time in Patent Office: 1,401 Days
• Field of Search: None
• US Class Current: 382/203
• CPC Class Codes: G06F 16/5854 using shape and object relationship
• CPC Class Codes: G06V 10/757 Matching configurations of points or features

Abstract

A system and a method for near-duplicate image detection performed by a physical computing system includes applying a feature determining function to a number of images, a feature being defined by a geometric shape, comparing characteristics of said geometric shapes defining said features from at least two of said number of images, and characterizing said at least two of said number of images as a near-duplicate match if a predetermined percentage of said features of said at least two images match.

Claims

1. A method for near-duplicate image detection performed by a physical computing system, the method comprising:

applying a feature detection function to a number of images individually to identify a number of features within each image, a feature being a region within a particular image defined by a geometric shape to identify the feature;

comparing characteristics of said geometric shapes identifying said features within a first image with characteristics of said geometric shapes identifying said features within a second image; and

characterizing said first and second images as a near-duplicate match if a predetermined percentage of said features of said first and second images match,

wherein the geometric shapes are not included within the content of the images.

2. The method of claim 1, in which said geometric shape comprises an ellipse.

3. The method of claim 2, in which said characteristics of said geometric shapes comprise a horizontal coordinate, a vertical coordinate, a major axis, a minor axis, an angle of rotation, or combinations thereof.

4. The method of claim 1, further comprising using a Gaussian distribution to determine a tolerance range, said tolerance range used to determine if said corresponding features from said first and second images match.

5. The method of claim 4, in which a standard deviation used by said Gaussian distribution is a value of a characteristic of said geometric shape multiplied by a coefficient.

6. The method of claim 5, further comprising adjusting said coefficient to adjust said tolerance range.

7. The method of claim 1, further comprising determining if said features from said first and second images match by determining if neighboring features of said features from said first and second images match.

8. A physical computing system comprising:

a processor; and

a memory communicatively coupled to said processor;

in which said processor;

applies a feature detection function to a number of images individually to identify a number of features within each image, a feature being a region within a particular image defined by a geometric shape to identify the feature;

compares characteristics of said geometric shapes; and

characterizes at least two of said number of images as near-duplicate matches if a predetermined percentage of said features of said at least two of said number of images match,

wherein the geometric shapes are not included within the content of the images.

9. The system of claim 8, in which said geometric shape comprises an ellipse.

10. The system of claim 9, in which said characteristics of said geometric shapes comprise a horizontal coordinate, a vertical coordinate, a major axis, a minor axis, an angle of rotation, or combinations thereof.

11. The system of claim 8, in which said processor uses a Gaussian distribution to determine a tolerance range, said tolerance range used to determine if a feature from one of said number of images matches a feature from another of said number of images.

12. The system of claim 11, in which a standard deviation used by said Gaussian distribution is a value of a characteristic of said geometric shape multiplied by a constant.

13. The system of claim 12, in which said processor adjusts said constant to adjust said tolerance range.

14. The system of claim 8, in which said processor determines if a feature from one of said number of images matches a feature from another of said number of images matches by determining if neighboring features of said features match.

15. A method for near-duplicate image detection performed by a physical computing system, the method comprising:

applying a feature detection function to a number of images individually to identify a number of features within each image, a feature being a region within a particular image defined by an ellipse to identify the feature;

comparing characteristics of said ellipses; and

characterizing at least two of said number of images as near-duplicate matches if a predetermined percentage of said features of said at least two of said number of images match;

in which given features are classified as a match if values of said characteristics of said ellipses defining the given features are within a predefined tolerance range,

wherein the ellipses are not included within the content of the images.