PARAMETERIZATION OF A CATEGORIZER FOR ADJUSTING IMAGE CATEGORIZATION AND RETRIEVAL

• US 20120054658A1
• Filed: 08/30/2010
• Published: 03/01/2012
• Est. Priority Date: 08/30/2010
• Status: Active Grant

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Chuat, Mathieu, Devin, Vincent
• Granted Patent: US 8,566,746 B2
• Time in Patent Office: Days
• US Class Current: 715/771
• CPC Class Codes: G06F 16/583 using metadata automatically derived from the content

Abstract

A system and method for digital object categorization or retrieval are provided. The method includes providing for a selector to be graphically presented to a user. The selector is variably adjustable within a range, by the user, to adjust a level of noise in at least one of digital object categorization and digital object retrieval. The range of the selector is normalized over a set of digital object categories based on scores output by a trained categorizer for each of a set of labeled test objects for each of the set of digital object categories and a label for each of the test objects. Selector position information is received by the system and at least one of digital object categorization and digital object retrieval is performed, based on the selector position information. One or more of the method steps may be performed with a computer processor.

Claims

1. A method for digital object categorization or retrieval comprising:

providing for graphically presenting a selector to a user, the selector including a position which is variably adjustable within a range by the user to adjust a level of noise in at least one of digital object categorization and digital object retrieval, the range of the selector having been normalized over a set of digital object categories based on scores output by a trained categorizer for each of a set of labeled test objects for each of the set of digital object categories and a label for each of the test objects;

receiving selector position information;

with a computer processor, performing at least one of digital object categorization and digital object retrieval, and

providing for outputting results of the at least one of digital object categorization and digital object retrieval, based on the selector position information.

2. The method of claim 1, wherein the selector is presented in a graphical user interface.

3. The method of claim 2, wherein the selector is displayed on a screen and is adjustable with a cursor control device.

4. The method of claim 1, wherein the selector is adjustable in the range between minimum and maximum values which, for each category, correspond to threshold scores within a range of possible scores output by the categorizer.

5. The method of claim 4, further comprising:

for each of the set of categories, identifying the threshold scores based on a computation of at least one of a recall rate, a precision rate, and a false positive rate for each of a set of possible scores output by the categorizer.

6. The method of claim 5, wherein a first threshold score is identified, for each category in the set of categories, based on a predetermined false positive rate which is the same for all categories in the set of categories.

7. The method of claim 4, wherein a lower one of the threshold scores is greater than a minimum of the possible scores output by the categorizer and a higher of the threshold scores is less than a maximum of the possible scores output by the categorizer.

8. The method of claim 6, wherein a second threshold score is based on at least one of a difference in precision rate and a difference in recall rate between the first and second thresholds, which difference is the same for all categories in the set of categories.

9. The method of claim 1, wherein the digital object comprises an image.

10. The method of claim 1, wherein when the method includes performing digital object categorization, the selector position variably adjusts a number of labels displayed or applied to a submitted digital object, each label corresponding to a respective one of the set of digital object categories.

11. The method of claim 10, wherein the outputting of the results includes displaying the labels applied the submitted digital object.

12. The method of claim 11, wherein the number of the displayed labels varies as the selector position is adjusted by a user.

13. The method of claim 11, wherein at least one of the displayed labels fades as the selector position is adjusted by a user.

14. The method of claim 1, wherein the method includes performing digital object categorization, the method further comprising computing a confidence level for a label applied to the submitted digital object including projecting a score for the digital object onto a plot of precision rate vs. category score output by the trained categorizer.

15. The method of claim 1, wherein when the method includes performing digital object retrieval, the selector position variably adjusts a number of digital objects retrieved or displayed in response to a submitted category-based query.

16. The method of claim 15, wherein the outputting of results comprises displaying the retrieved digital objects.

17. The method of claim 16, wherein the number of the displayed digital objects varies as the selector position is adjusted by a user.

18. The method of claim 1, further comprising computing a performance for each category based on the scores output by a trained categorizer for each of the set of labeled test objects and their labels.

19. The method of claim 18, wherein when the method includes digital object categorization, the labels are ranked by category performance.

20. A computer program product comprising a non-transitory recording medium encoding instructions which, when executed by a computer, perform the method of claim 1.

21. A system comprising memory which stores instructions for performing the method of claim 1 and at least one processor, in communication with the memory, for executing the instructions.

22. A system for variably adjusting a level of noise for digital object categorization or retrieval comprising:

memory storing instructions for presenting a selector with a position which is variably adjustable by a user to adjust a level of noise in at least one of digital object categorization and digital object retrieval, an impact of the selector on an output of the digital object categorization or digital object retrieval being normalized over a set of digital object categories based on scores output by a trained categorizer for each of a set of labeled test objects for each of the set of digital object categories and a label for each of the test objects; and

a computer processor which executes the instructions.

23. The system of claim 22, further comprising the digital object categorizer stored in memory which labels an input digital object, based on the selector position.

24. A method for assisting a user in variably adjusting a level of noise for digital object categorization or retrieval comprising:

submitting a set of labeled test images to a probabilistic categorizer comprising a categorizer model for each of a set of categories to generate scores for each of the labeled test images for each of the categories in the set;

for each of the set of categories, computing threshold scores based on the generated scores for that category and labels of the test images;

presenting a selector in a graphical user interface with a position which is variably adjustable by a user to adjust the level of noise in at least one of digital object categorization and digital object retrieval, the selector having been normalized over the set of digital object categories based on the computed threshold scores.