Method and system for denoising pairs of mutually interfering signals

• US 7,623,725 B2
• Filed: 10/14/2005
• Issued: 11/24/2009
• Est. Priority Date: 10/14/2005
• Status: Expired due to Fees

• Current Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Original Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Inventors: Ordentlich, Erik, Seroussi, Gadiel
• Primary Examiner(s): Bali, Vikkram
• Assistant Examiner(s): ENTEZARI, MICHELLE M
• Application Number:
  US11/250,831
• Publication Number:
  US 20070086058A1
• Time in Patent Office: 1,502 Days
• Field of Search: None
• US Class Current: 382/254
• CPC Class Codes: G06V 10/30 Noise filtering

Abstract

In various embodiments of the present invention, a number n of mutually interfering signals are denoised by selecting a discrete universal denoiser method that denoises n mutually interfering signals, tuning the discrete universal denoiser to denoise the n mutually interfering signals, and denoising the n mutually interfering signals by applying the tuned discrete universal denoiser to the n mutually interfering signals.

Claims

1. A method for denoising n equals two or more mutually interfering signals, the method comprising:

using a processor;

selecting a discrete universal denoiser method that denoises n mutually interfering signals;

tuning the discrete universal denoiser to denoise the n mutually interfering signals; and

denoising the n mutually interfering signals by applying the tuned discrete universal denoiser to the n mutually interfering signals.

2. The method of claim 1 wherein the selected discrete universal denoiser denoises a signal by:

determining symbol-transition probabilities for the noise-introducing signal-acquisition process;

determining a measure of distortion produced with respect to a clean signal corresponding to the noisy signal by substituting a given replacement symbol for a given clean symbol;

counting occurrences of symbols within contexts in the noisy signal; and

replacing symbols in the noisy signal with replacement symbols that provide a smallest estimated distortion with respect to the clean signal corresponding to the noisy signal.

3. The method of claim 1 wherein tuning the discrete universal denoiser to denoise the n mutually interfering signals further includes:

receiving n pairs of noisy and corresponding clean signals;

selecting initial discrete universal denoiser parameters; and

iteratively

denoising the noisy signal composed of n mutually interfering signals to produce a corresponding denoised signal associated with the n mutually interfering signals,

comparing the denoised signal associated with n mutually interfering signals to the clean signals, and

adjusting discrete universal denoiser parameters.

4. The method of claim 3 wherein the iterative denoising, comparing, and adjusting steps are carried out until the differences between the denoised signal associated with each set of n mutually interfering signals and the clean signals associated with the set of n mutually interfering signals meet one or more predetermined criteria, the one or more predetermined criteria including one or more of:

the differences falling below a threshold value;

the differences representing a local minimum; and

the differences representing a global minimum.

5. The method of claim 3 wherein the iterative denoising, comparing, and adjusting steps are repeated until a maximum number of iterations have been carried out.

6. The method of claim 3 wherein the discrete universal denoiser parameters include:

a channel noise model Π

;

a distortion model;

a number of contexts;

sizes and geometries of the contexts,

a number of iterations in various DUDE-CTI iterative embodiments, and

a number of recursions in various DUDE-CTI recursive embodiments.

7. A system that includes a signal denoising component that denoises a set of n mutually interfering noisy signals obtained by a noise-introducing signal-acquisition process, the denoising component including:

an embodiment of a selected discrete universal denoiser method that denoises noisy signals, tuned for denoising n mutually interfering noisy signals.

8. A method for denoising images of double-sided-document pages obtained by a noise-introducing image-acquisition process, the method comprising:

using a processor;

selecting a discrete universal denoiser method that denoises noisy images;

tuning the discrete universal denoiser to denoise images of double-sided-document pages; and

denoising images of double-sided-document pages by applying the tuned discrete universal denoiser to the images of double-sided-document pages.

9. The method of claim 8 wherein the selected discrete universal denoiser denoises a noisy image by:

determining symbol-transition probabilities for the noise-introducing image-acquisition process;

determining a measure of distortion produced with respect to a clean image by substituting a given replacement symbol for a given clean symbol;

counting occurrences of symbols within contexts in the noisy image; and

replacing symbols in the noisy image with replacement symbols that provide a smallest estimated distortion with respect to the clean image corresponding to the noisy image.

10. The method of claim 8 wherein tuning the discrete universal denoiser to denoise images of double-sided-document pages further includes:

receiving pairs of noisy and corresponding clean images;

selecting initial discrete universal denoiser parameters; and

iteratively

denoising the noisy image of each pair of images to produce a corresponding denoised image associated with the pair,

comparing the denoised image associated with each pair of images to the clean image of the pair of images, and

adjusting discrete universal denoiser parameters.

11. The method of claim 10 wherein the iterative denoising, comparing, and adjusting steps are carried out until the differences between the denoised image associated with each pair of images and the clean image of the pair of images meet one or more predetermined criteria, the one or more predetermined criteria including one or more of:

the differences falling below a threshold value;

the differences representing a local minimum; and

the differences representing a global minimum.

12. The method of claim 10 wherein the iterative denoising, comparing, and adjusting steps are repeated until a maximum number of iterations have been carried out.

13. The method of claim 10 wherein the discrete universal denoiser parameters include:

a channel noise model Π

;

a distortion model;

a number of contexts;

sizes and geometries of the contexts,

a number of iterations in various DUDE-CTI iterative embodiments, and

a number of recursions in various DUDE-CTI recursive embodiments.

14. A system that includes an image denoising component that denoises images of double-sided-document pages obtained by a noise-introducing image-acquisition process, the denoising component including:

an embodiment of a selected discrete universal denoiser method that denoises noisy images, tuned for denoising double-sided-document pages.

15. A method for denoising images of double-sided-document pages obtained by a noise-introducing image-acquisition process, the method comprising:

using a processor;

selecting a discrete universal denoiser method that denoises noisy images;

tuning the discrete universal denoiser to denoise pairs of images, each pair comprising an image of a front-side and an image of a back-side of a double-sided-document page; and

pairwise denoising images of the front and back sides of double-sided-document pages by, for each double-sided-document page, aligning the image of the front-side and the image of the back-side of the double-sided-document page and applying the tuned discrete universal denoiser to a combined signal comprising the image of the front-side and the image of the back-side of the double-sided-document page.

16. The method of claim 15 wherein the selected discrete universal denoiser denoises a combined signal comprising the image of the front-side and the image of the back-side of the double-sided-document page by:

determining symbol-transition probabilities for the noise-introducing image-acquisition process that include a degree of bleed-through α

;

determining a measure of distortion produced with respect to a clean image by substituting a given replacement symbol for a given clean symbol;

counting occurrences of symbols within contexts in both the image of the front-side and the image of the back-side of the double-sided-document page; and

replacing symbols in the image of the front-side and the image of the back-side of the double-sided-document page with replacement symbols that provide a smallest estimated distortion with respect to the clean images corresponding to the noisy image of the front-side and the image of the back-side of the double-sided-document page.

17. The method of claim 15 wherein tuning the discrete universal denoiser to denoise double-sided-document pages further includes:

receiving quadruples of images, each quadruple including noisy images of the front-side and back-side of a double-sided-document page and corresponding clean images of the front-side and back-side of the double-sided-document page;

selecting initial discrete universal denoiser parameters; and

iteratively

aligning the image of the front-side and the image of the back-side of the double-sided-document page in each quadruple,

denoising the aligned noisy images of each quadruple of images by applying the discrete universal denoiser to a combined signal comprising the image of the front-side and the image of the back-side of the double-sided-document page to produce corresponding denoised images associated with the quadruple,

comparing the denoised images associated with each quadruple of images to the corresponding clean images of the quadruple of images, and

adjusting discrete universal denoiser parameters.

18. The method of claim 17 wherein the iterative aligning, denoising, comparing, and adjusting steps are carried out until the differences between the denoised images associated with each quadruple of images and the clean image of the pair of images meet one or more predetermined criteria, the one or more predetermined criteria including one or more of:

the differences falling below a threshold value;

the differences representing a local minimum; and

the differences representing a global minimum.

19. The method of claim 17 wherein the iterative aligning, denoising, comparing, and adjusting steps are repeated until a maximum number of iterations have been carried out.

20. The method of claim 17 wherein the discrete universal denoiser parameters include:

a channel noise model Π

, including the degree of bleed through α

;

a distortion model;

a number of contexts;

sizes and geometries of the contexts,

a number of iterations in various DUDE-CTI iterative embodiments, and

a number of recursions in various DUDE-CTI recursive embodiments.

21. A system that includes an image denoising component that pairwise denoises images of front and back sides of double-sided-document pages obtained by a noise-introducing image-acquisition process, the denoising component including:

an embodiment of a selected discrete universal denoiser method that denoises combined signals, each combined signal comprising the front-side and back-side noisy images of a double-sided-document page, the selected discrete universal denoiser method tuned for denoising double-sided-document pages.