Resource-aware computer vision

• US 8,811,743 B2
• Filed: 06/09/2010
• Issued: 08/19/2014
• Est. Priority Date: 06/09/2010
• Status: Expired due to Fees

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Kapoor, Ashish, Vijayanarasimhan, Sudheendra N.
• Primary Examiner(s): CONWAY, THOMAS A
• Application Number:
  US12/796,686
• Publication Number:
  US 20110305392A1
• Time in Patent Office: 1,532 Days
• Field of Search: None
• US Class Current: 382/190
• CPC Class Codes: G06F 18/211 Selection of the most significant subset of features
• CPC Class Codes: G06V 10/50 by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
• CPC Class Codes: G06V 10/771 Feature selection, e.g. selecting representative features from a multi-dimensional feature space

Abstract

The described implementations relate to computer vision. In one case image data is received. Resource constraints associated with processing the image data are evaluated. Further, a visual recognition and detection processing strategy is selected for the image data based at least in part on the evaluated resource constraints.

Claims

1. A system, comprising:

a feature extraction component configured to receive an image and to identify parts of the image for further evaluation;

a pattern recognition component configured to recognize object portions from the image parts;

a parameter-based decision component configured to;

evaluate both a current power constraint of the system and a value-of-information for at least one of the image parts, the value-of-information being influenced by the current power constraint, and

determine whether to influence the feature extraction component and the pattern recognition component based upon the evaluation of the current power constraint and the value-of-information; and

,

a processor that executes computer-executable instructions associated with at least one of the feature extraction component, the pattern recognition component, or the parameter-based decision component.

2. The system of claim 1, wherein the pattern recognition component is further configured to identify a feature that is useful for recognizing an individual object portion.

3. The system of claim 1, wherein the parameter-based decision component is configured to obtain the value-of-information in relation to processing cycles of the feature extraction component and the pattern recognition component.

4. The system of claim 1, further comprising a set of training images that is accessible by the feature extraction component and the pattern recognition component and that relates individual features that are indicative of individual object portions.

5. The system of claim 1, wherein the parameter-based decision component is configured to stop the feature extraction component and the pattern recognition component before completion based on a diminishing returns value that is derived from the value-of-information.

6. The system of claim 1, wherein the parameter-based decision component is further configured to cause the feature extraction component and the pattern recognition component to implement an anytime solution that is influenced by the current power constraint of the system.

7. The system of claim 6, wherein the anytime solution includes stopping the feature extraction component and the pattern recognition component before completion.

8. The system of claim 1, wherein the parameter-based decision component is further configured to monitor progress of the feature extraction component and the pattern recognition component in individual processing cycles, the progress related to the value-of-information.

9. The system of claim 8, wherein the parameter-based decision component is further configured to influence the feature extraction component and the pattern recognition component to stop processing and generate an output when the progress produced in at least one of the individual processing cycles falls below a threshold.

10. The system of claim 8, wherein the parameter-based decision component is further configured to influence the feature extraction component and the pattern recognition component to stop processing and generate an output when an average amount of the progress produced in a number of the individual processing cycles falls below a threshold.

11. The system of claim 1, wherein the parameter-based decision component is further configured to allot an amount of computation to the feature extraction component and the pattern recognition component based upon the evaluation.

12. The system of claim 2, wherein the feature extraction component is configured to map the image utilizing a grid model to identify the image parts as grid parts.

13. The system of claim 12, wherein the feature extraction component is configured to access training data that identifies the feature that is indicative of the individual object portion contained in an individual grid part.

14. The system of claim 1, wherein the feature extraction component is configured to map the image utilizing a grid model to identify the image parts as grid parts and to compute the value-of-information for at least one grid part.

15. The system of claim 14, wherein the feature extraction component is further configured to compute another value-of-information for at least one local feature in the at least one grid part.

16. The system of claim 15, wherein the parameter-based decision component influences the feature extraction component to return a partial answer based on the value-of-information or the another value-of-information.

17. A system, comprising:

a feature extraction component configured to receive an image and to identify parts of the image for further evaluation;

a pattern recognition component configured to recognize object portions from the image parts;

a parameter-based decision component configured to;

calculate multiple value-of-information values in relation to processing cycles, wherein an individual value-of-information value is a ratio of gain in scores of a feature to computational cost of the feature,

compare at least some of the multiple value-of-information values to determine a diminishing returns value, and

determine whether to influence the feature extraction component and the pattern recognition component by comparing the diminishing returns value to a predefined threshold; and

,

a processor configured to execute computer-executable instructions associated with at least one of the feature extraction component, the pattern recognition component, or the parameter-based decision component.

18. The system of claim 17, wherein the parameter-based decision component is further configured to:

compare a first value-of-information value calculated after a first processing cycle to a second value-of-information value calculated after a second processing cycle, yielding a first progress,

compare the second value-of-information value to a third value-of-information value calculated after a third processing cycle, yielding a second progress,

determine the diminishing returns value by comparing the first progress and the second progress, and

influence the feature extraction component and the pattern recognition component to stop processing and generate an output when the diminishing returns value reaches the predefined threshold.

19. A system, comprising:

a feature extraction component configured to receive an image and to identify feature-location pairs in the image for further evaluation, an individual feature-location pair including an individual feature and a corresponding location;

a pattern recognition component configured to recognize object portions from features of the feature-location pairs;

a parameter-based decision component configured to;

evaluate constraints of computational resources of the system,

calculate value-of-information values for the feature-location pairs, wherein an individual value-of-information value is a function of both gain in scores associated with the individual feature at the corresponding location and computational cost related to the computational resources, and

determine whether to influence the feature extraction component and the pattern recognition component based upon the constraints of the computational resources and the value-of-information values; and

,

a processor configured to execute computer-executable instructions associated with at least one of the feature extraction component, the pattern recognition component, or the parameter-based decision component.

20. The system of claim 19, wherein the parameter-based decision component is further configured to stop processing of the image and generate an output based upon the constraints of the computational resources.