Methods and Systems for Image Noise Processing

• US 20080152247A1
• Filed: 12/21/2006
• Published: 06/26/2008
• Est. Priority Date: 12/21/2006
• Status: Active Grant

• Current Assignee: Sharp Kabushiki Kaisha (Hon Hai Precision Industry Co., Ltd.)
• Original Assignee: Sharp Kabushiki Kaisha (Hon Hai Precision Industry Co., Ltd.)
• Inventors: Sun, Shijun, Oh, Byung Tae
• Granted Patent: US 7,742,652 B2
• Time in Patent Office: Days
• US Class Current: 382/260
• CPC Class Codes: G06T 2207/10024 Color image
• CPC Class Codes: G06T 2207/20012 Locally adaptive
• CPC Class Codes: G06T 2207/20192 Edge enhancement; Edge preservation
• CPC Class Codes: G06T 5/70 Denoising; Smoothing
• CPC Class Codes: G06V 10/30 Noise filtering
• CPC Class Codes: H04N 19/85 using pre-processing or post-processing specially adapted for video compression

Abstract

Embodiments of the present invention comprise systems and methods for identifying and separating noise from an image.

Claims

1. A method for image noise separation, said method comprising:

a) processing an original input image to create a first-layer LH filtered image, LH₁, which has been low-pass filtered in a first direction and high-pass filtered in a second direction;

b) processing said input image to create a first-layer HL filtered image, HL₁, which has been high-pass filtered in said first direction and low-pass filtered in said second direction;

c) processing said input image to create a first-layer LL filtered image, LL₁, which has been low-pass filtered in said first direction and low-pass filtered in said second direction;

d) combining said first-layer LH filtered image and said first-layer HL filtered image to produce a first-layer edge energy feature, EE₁;

e) repeating steps b-d using said first-layer LL filtered image as an input image thereby creating a second-layer LH filtered image, LH₂, a second-layer HL filtered image, HL₂, and a second-layer LL filtered image, LL₂;

f) combining said second-layer LH filtered image and said second-layer HL filtered image to produce a second-layer edge energy feature, EE₂;

g) creating an edge map, EM, from at least one of said edge energy features; and

h) creating a de-noised image by modifying values in said original input image with values in at least one of said LL filtered images at locations that are not designated as edge areas by said edge map.

2. A method as described in claim 1 further comprising:

a) repeating steps b-d using said second-layer LL filtered image, LL₂, as an input image thereby creating a third-layer LH filtered image, LH₃, a third-layer HL filtered image, HL₃, and a third-layer LL filtered image, LL₃;

b) repeating steps b-d using said third-layer LL filtered image, LL₃, as an input image thereby creating a fourth-layer LH filtered image, LH₄, a fourth-layer HL filtered image, HL₄, and a fourth-layer LL filtered image, LL₄;

c) combining said third-layer LH filtered image and said third-layer HL filtered image to produce a third-layer edge energy feature, EE₃;

d) combining said fourth-layer LH filtered image and said fourth-layer HL filtered image to produce a fourth-layer edge energy feature, EE₄;

e) creating a combined energy feature by averaging a plurality of said edge energy features; and

f) wherein said creating an edge map comprises using a two-threshold, soft threshold process on said combined energy feature.

3. A method as described in claim 2 wherein said creating a de-noised image comprises application of the following equation:

Denoise(i, j)=LL₃(i, j)+EM(i, j)×

[I(i, j)−

LL₃(i, j)].

4. A method as described in claim 2 wherein said creating a de-noised image comprises application of the following equation:

Denoise=LL₃·

(1−

EM)²+LL₁·

EM(1−

EM)+I·

EM

5. A method as described in claim 2 wherein said creating a combined energy feature comprises averaging said EE₃ and said EE₄.

6. A method for image noise separation, said method comprising:

a) receiving a color input image;

b) performing multi-layer decomposition on each color channel of said color input image thereby producing at least one edge energy feature for each color channel and a first-layer low-pass filtered image and a third-layer low-pass filtered image for each color channel;

c) representing said edge energy features with coefficients;

d) weighting the color channels of said color input image with said coefficients to produce a combined image;

e) performing multi-layer decomposition on said combined image to produce at least one combined image edge energy feature and at least one combined image low-pass filtered image;

f) creating an edge map based on said combined image edge energy feature; and

g) creating a de-noised color image by modifying values in each color channel of said color input image with values from said low-pass filtered color channel images at locations that are not designated as edge areas by said edge map.

7. A method as described in claim 6 wherein said representing said edge energy features with coefficients comprises a principle component analysis process.

8. A method as described in claim 6 wherein said combined image edge energy is determined by averaging a plurality of edge energy features for different decomposition layers.

9. A method as described in claim 6 wherein said combined image edge energy is determined by averaging a third-layer edge energy feature and a fourth-layer edge energy feature for said combined image.

10. A method as described in claim 6 wherein said performing multi-layer decomposition on said combined image produces a third-layer combined-image low-pass filtered image.

11. A method as described in claim 6 wherein said creating an edge map comprises using a two-threshold, soft threshold process on said combined image edge energy feature.

12. A method as described in claim 6 wherein said creating a de-noised color image comprises application of the following equation for each color channel:

Denoise(i, j)=LL₃(i, j)+EM(i, j)×

[I(i, j)−

LL₃(i, j)]

wherein I represents an input image value, EM represent edge map values and LL₃ represents a third-layer, combined-image, low-pass filtered image value.

13. A method as described in claim 6 wherein said creating a de-noised color image comprises application of the following equation for each color channel:

Denoise=LL₃·

(1−

EM)²+LL₁·

EM(1−

EM)+I·

EM

wherein I represents an input image value, EM represent edge map values, LL₃ represents a third-layer, combined-image, low-pass filtered image value and, LL₁ represents a first-layer, combined-image, low-pass filtered image value.

14. A method as described in claim 6 further comprising creating a noise image.

15. A system for image noise separation, said method comprising:

a) a color-channel multi-layer filter processor for performing multi-layer decomposition on each color channel of a color input image thereby producing at least one edge energy feature for each color channel, a first-layer low-pass filtered image and a third-layer low-pass filtered image for each color channel of said color input image;

b) a coefficient generator for representing said edge energy features with coefficients;

c) a linear projection processor for weighting the color channels of said color input image with said coefficients to produce a combined image;

d) a combined-image, multi-layer filter processor for performing multi-layer decomposition on said combined image to produce at least one combined image edge energy feature and at least one combined image low-pass filtered image;

e) an edge map processor for creating an edge map based on said combined image edge energy feature; and

f) a de-noise processor for creating a de-noised color image by modifying values in each color channel of said color input image with values from said low-pass filtered color channel images at locations that are not designated as edge areas by said edge map.

16. A system as described in claim 15 wherein said coefficient generator for representing said edge energy features with coefficients uses a principle component analysis process.

17. A method as described in claim 15 wherein combined-image, multi-layer filter processor averages a third-layer edge energy feature and a fourth-layer edge energy feature to determine said combined image edge energy feature.

18. A method as described in claim 15 wherein said combined-image, multi-layer filter processor for performing multi-layer decomposition on said combined image produces a third-layer, combined-image, low-pass filtered image.

19. A method as described in claim 15 wherein said edge map processor for creating an edge map uses a two-threshold, soft threshold process on said combined image edge energy feature.

20. A method as described in claim 15 wherein said de-noise processor for creating a de-noised color image applies the following equation for each color channel:

Denoise=LL₃·

(1−

EM)²+LL₁·

EM(1−

EM)+I·

EM

wherein I represents an input image value, EM represent edge map values, LL₃ represents a third-layer, combined-image, low-pass filtered image value and, LL₁ represents a first-layer, combined-image, low-pass filtered image value.