Noise reduction with classification-based contrast preservation

Patent metadata

Field	Value
Publication number	US-11200641-B1
Application number	US-201916580501-A
Country	US
Kind code	B1
Filing date	Sep 24, 2019
Priority date	Sep 27, 2018
Publication date	Dec 14, 2021
Grant date	Dec 14, 2021

Abstract

In one embodiment, a method includes obtaining an image comprising a plurality of pixels. The method includes determining, for a particular pixel of the plurality of pixels, a feature value. The method includes selecting, based on the feature value, a set of selected pixels from a set of candidate pixels in an image region surrounding the particular pixel. The method includes denoising the particular pixel based on the set of selected pixels.

First claim

What is claimed is:

1. A method comprising:
obtaining an image comprising a plurality of pixels;
determining, for a particular pixel of the plurality of pixels, a feature value that indicates on which side of an edge the particular pixel is located;
determining, for the particular pixel, an adjusted pixel value based on a pixel value of the particular pixel and the feature value of the particular pixel;
selecting, based on the feature value, a set of selected pixels from a set of candidate pixels in an image region surrounding the particular pixel in response to determining that a difference between pixel values of the set of selected pixels and the adjusted pixel value is less than a threshold; and
denoising the particular pixel based on the set of selected pixels.

2. The method of claim 1 , wherein determining the feature value comprises determining a Laplacian of Gaussian (LoG) feature value by convolving an LoG kernel with an image region surrounding the pixel.

3. The method of claim 1 , wherein the threshold is based on a noise value of the particular pixel derived from a noise model.

4. The method of claim 1 , wherein determining the adjusted pixel value includes adding the pixel value of the particular pixel to the product of the feature value of the particular pixel and a tuning parameter.

5. The method of claim 4 , wherein the tuning parameter is based on the feature value.

6. The method of claim 5 , wherein the tuning parameter is selected, based on a quantization of the feature value, from a plurality of predefined tuning parameters.

7. The method of claim 1 , wherein denoising the particular pixel based on the set of selected pixels includes averaging the set of selected pixels.

8. The method of claim 7 , wherein averaging the set of selected pixels includes determining a weighted average of pixel values of the set of selected pixels.

9. The method of claim 1 , further comprising:
generating a denoised image based on denoising the particular pixel;
decomposing the denoised image into a low-frequency component and a high-frequency component;
generating a detail-modulation image based on the high-frequency component and local variances of the high-frequency component; and
generating a post-processed image by adding, to the denoised image, a scaled version of the detail-modulation image.

10. A non-transitory computer-readable medium having instructions encoded thereon that, when executed by one or more processors of an electronic device, cause the electronic device to perform a method comprising:
obtaining an image comprising a plurality of pixels;
determining, for a particular pixel of the plurality of pixels, a feature value that indicates on which side of an edge the particular pixel is located;
determining, for the particular pixel, an adjusted pixel value based on a pixel value of the particular pixel and the feature value of the particular pixel;
selecting, based on the feature value, a set of selected pixels from a set of candidate pixels in an image region surrounding the particular pixel in response to determining that a difference between pixel values of the set of selected pixels and the adjusted pixel value is less than a threshold; and
denoising the particular pixel based on the set of selected pixels.

11. The non-transitory computer-readable medium of claim 10 , wherein determining the feature value comprises determining a Laplacian of Gaussian (LoG) feature value by convolving an LoG kernel with an image region surrounding the pixel.

12. The non-transitory computer-readable medium of claim 10 , wherein the threshold is based on a noise value of the particular pixel derived from a noise model.

13. The non-transitory computer-readable medium of claim 10 , wherein determining the adjusted pixel value includes adding the pixel value of the particular pixel to the product of the feature value of the particular pixel and a tuning parameter.

14. The non-transitory computer-readable medium of claim 13 , wherein the tuning parameter is based on the feature value.

15. The non-transitory computer-readable medium of claim 14 , wherein the tuning parameter is selected, based on a quantization of the feature value, from a plurality of predefined tuning parameters.

16. The non-transitory computer-readable medium of claim 10 , wherein denoising the particular pixel based on the set of selected pixels includes averaging the set of selected pixels.

17. The non-transitory computer-readable medium of claim 10 , wherein the method further comprises:
generating a denoised image based on denoising the particular pixel;
decomposing the denoised image into a low-frequency component and a high-frequency component;
generating a detail-modulation image based on the high-frequency component and local variances of the high-frequency component; and
generating a post-processed image by adding, to the denoised image, a scaled version of the detail-modulation image.

18. A system comprising:
a non-transitory memory; and
one or more processors configured to:
obtain an image comprising a plurality of pixels;
determine, for a particular pixel of the plurality of pixels, a feature value that indicates on which side of an edge the particular pixel is located;
determine, for the particular pixel, an adjusted pixel value based on a pixel value of the particular pixel and the feature value of the particular pixel;
select, based on the feature value, a set of selected pixels from a set of candidate pixels in an image region surrounding the particular pixel in response to determining that a difference between pixel values of the set of selected pixels and the adjusted pixel value is less than a threshold; and
denoise the particular pixel based on the set of selected pixels.

19. The method of claim 1 , wherein the selecting comprises selecting the set of selected pixels such that the set of selected pixels includes a subset of the set of candidate pixels that is on the same side of the edge as the particular pixel.

Assignees

• Apple Inc

Inventors

• Baqai Farhan A
• Cheng Wu

Classifications

• G06T2207/20192

  Edge enhancement; Edge preservation · CPC title

• G06T2207/10024

  Color image · CPC title

• G06T5/20

  using local operators · CPC title

• H04N1/60

  Colour correction or control {(H04N1/54 takes precedence)} · CPC title

• H04N1/409

  Edge or detail enhancement; Noise or error suppression · CPC title