Building image-based models by mapping non-linear optmization to streaming architectures

• US 20040100473A1
• Filed: 11/22/2002
• Published: 05/27/2004
• Est. Priority Date: 11/22/2002
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Grzeszczuk, Radek, Molinov, Sergey A., Hillesland, Karl E.
• Granted Patent: US 6,999,074 B2
• Time in Patent Office: Days
• US Class Current: 345/581
• CPC Class Codes: G06T 15/50 Lighting effects
• CPC Class Codes: G06T 17/10 Constructive solid geometry [CSG] using solid primitives, e.g. cylinders, cubes

Abstract

Building an image-based model may be accomplished by defining an approximation set of functions for building a representation of the image-based model, defining an error function that quantifies how well the approximation set reproduces image data of the image-based model, and adjusting parameters of the approximation set to produce a best approximation to the image data.

Claims

1. A method comprising:

defining an approximation set of functions for building a representation of an image-based model;

defining an error function that quantifies how well the approximation set reproduces image data of the image-based model; and

adjusting parameters of the approximation set to produce a best approximation to the image data.

2. The method of claim 1, further comprising minimizing an error metric produced by the error function and adjusting the parameters according to the error metric to produce the best approximation to the image data.

3. An article comprising:

a storage medium having a plurality of machine accessible instructions, wherein when the instructions are executed by a processor, the instructions provide for building image-based models by defining an approximation set of functions for building a representation of an image-based model, defining an error function that quantifies how well the approximation set reproduces image data of the image-based model, and adjusting parameters of the approximation set to produce a best approximation to the image data.

4. The article of claim 3, further comprising instructions for minimizing an error metric produced by the error function and adjusting the parameters according to the error metric to produce the best approximation to the image data.

5. A method comprising:

partitioning samples of an image and parameters of an approximation set of functions for an image-based model;

determining dependency of the image samples on the approximation set parameters;

defining an error function and data structures for computation of error derivatives using the image samples and approximation set parameters; and

optimizing the approximation set parameters using the error function to produce a best approximation to the image.

6. The method of claim 5, further comprising removing redundant image samples.

7. The method of claim 5, wherein optimizing the approximation set parameters comprises applying a steepest descent algorithm.

8. The method of claim 5, wherein optimizing the approximation set parameters comprises applying a Newton method algorithm.

9. The method of claim 5, wherein optimizing the approximation set parameters comprises applying a conjugate gradient algorithm.

10. The method of claim 5, further comprising minimizing an error metric produced by the error function and adjusting the approximation set parameters according to the error metric to produce the best approximation to the image.

11. An article comprising:

a storage medium having a plurality of machine accessible instructions, wherein when the instructions are executed by a processor, the instructions provide for building image-based models by partitioning samples of an image and parameters of an approximation set of functions for an image-based model, determining dependency of the image samples on the approximation set parameters, defining an error function and data structures for computation of error derivatives using the image samples and approximation set parameters, and optimizing the approximation set parameters using the error function to produce a best approximation to the image.

12. The article of claim 11, further comprising removing redundant image samples.

13. The article of claim 11, wherein optimizing the approximation set parameters comprises applying a steepest descent algorithm.

14. The article of claim 11, wherein optimizing the approximation set parameters comprises applying a Newton method algorithm.

15. The article of claim 11, wherein optimizing the approximation set parameters comprises applying a conjugate gradient algorithm.

16. The method of claim 11, further comprising minimizing an error metric produced by the error function and adjusting the approximation set parameters according to the error metric to produce the best approximation to the image.

17. A system for building image-based models comprising:

a first component to partition samples of an image and parameters of an approximation set of functions, and to determine dependency of image samples on the approximation set parameters;

a second component to define an error function and data structures for computation of error derivatives using the image samples and the approximation set parameters; and

a third component to optimize the approximation set parameters using the error function.

18. The system of claim 17, wherein the first component comprises a fourth component to remove redundant image samples.

19. The system of claim 17, wherein the third component optimizes the approximation set parameters by applying a steepest descent algorithm.

20. The system of claim 17, wherein the third component optimizes the approximation set parameters by applying a Newton method algorithm.

21. The system of claim 17, wherein the third component optimizes the approximation set parameters by applying a conjugate gradient algorithm.

22. The system of claim 17, wherein the third component minimizes an error metric produced by the error function and adjusts the approximation set parameters according to the error metric to produce the best approximation to the image.