System and method for adaptive discrete cosine transform (DCT) noise filtering for digital images

Patent metadata

Field	Value
Publication number	US-12367550-B2
Application number	US-202217930335-A
Country	US
Kind code	B2
Filing date	Sep 7, 2022
Priority date	Apr 29, 2022
Publication date	Jul 22, 2025
Grant date	Jul 22, 2025

Abstract

A method includes dividing an image into overlapping image patches each having a specified size. The method also includes analyzing content of each image patch using a mathematical transform technique to classify each image patch into at least one class. The method further includes filtering each image patch for noise suppression by suppressing one or more transform coefficients of the image patch. An amount of suppression for each of the one or more transform coefficients is selected according to the at least one class of the image patch. In addition, the method includes reconstructing the filtered image patches into an output image.

First claim

What is claimed is:

1. A method comprising:
dividing an image into overlapping image patches each having a specified size;
analyzing content of each image patch using a mathematical transform technique to classify each image patch into at least one class, wherein the at least one class includes at least one of: a dark patch, a bright patch, or a patch having strong edge content;
filtering each image patch for noise suppression by suppressing one or more transform coefficients of the image patch, wherein an amount of suppression for each of the one or more transform coefficients is selected according to the at least one class of the image patch such that (i) the amount of suppression for image patches classified as dark patches is increased and (ii) the amount of suppression for image patches classified as having strong edge content is reduced; and
reconstructing the filtered image patches into an output image.

2. The method of claim 1 , further comprising:
before dividing the image into the overlapping image patches:
performing a gamma correction on the image; and
converting the image from a red-green-blue (RGB) domain to a luma-chroma (YUV) domain.

3. The method of claim 2 , wherein analyzing the content of each image patch comprises:
determining multiple transform coefficients of the image patch, the multiple transform coefficients comprising the one or more suppressed transform coefficients;
determining multiple luma values of the image patch based on the multiple transform coefficients of the image patch; and
classifying the image patch as a dark patch or a bright patch based on the determined luma values.

4. The method of claim 3 , wherein analyzing the content of each image patch further comprises:
determining edge content of the image patch based on the multiple transform coefficients of the image patch; and
classifying the image patch as a patch with strong edge content or a patch without strong edge content based on the determined edge content.

5. The method of claim 4 , wherein filtering each image patch comprises:
filtering the image patch based on the determined luma values and the determined edge content of the image patch.

6. The method of claim 2 , wherein analyzing the content of each image patch comprises:
using a trained machine learning model to predict the at least one class for the image patch.

7. The method of claim 1 , wherein filtering each image patch comprises:
determining one or more dominant edges in the image patch; and
wherein the one or more transform coefficients of the image patch are suppressed while preserving one or more other spectral coefficients relevant to the one or more dominant edges.

8. An electronic device comprising:
at least one processor configured to:
divide an image into overlapping image patches each having a specified size;
analyze content of each image patch using a mathematical transform technique to classify each image patch into at least one class, wherein the at least one class includes at least one of: a dark patch, a bright patch, or a patch having strong edge content;
filter each image patch for noise suppression by suppressing one or more transform coefficients of the image patch, wherein an amount of suppression for each of the one or more transform coefficients is selected according to the at least one class of the image patch such that (i) the amount of suppression for image patches classified as dark patches is increased and (ii) the amount of suppression for image patches classified as having strong edge content is reduced; and
reconstruct the filtered image patches into an output image.

9. The electronic device of claim 8 , wherein the at least one processor is further configured, before dividing the image into the overlapping image patches, to:
perform a gamma correction on the image; and
convert the image from a red-green-blue (RGB) domain to a luma-chroma (YUV) domain.

10. The electronic device of claim 9 , wherein, to analyze the content of each image patch, the at least one processor is configured to:
determine multiple transform coefficients of the image patch, the multiple transform coefficients comprising the one or more suppressed transform coefficients;
determine multiple luma values of the image patch based on the multiple transform coefficients of the image patch; and
classify the image patch as a dark patch or a bright patch based on the determined luma values.

11. The electronic device of claim 10 , wherein, to analyze the content of each image patch, the at least one processor is further configured to:
determine edge content of the image patch based on the multiple transform coefficients of the image patch; and
classify the image patch as a patch with strong edge content or a patch without strong edge content based on the determined edge content.

12. The electronic device of claim 11 , wherein, to filter each image patch, the at least one processor is configured to filter the image patch based on the determined luma values and the determined edge content of the image patch.

13. The electronic device of claim 9 , wherein, to analyze the content of each image patch, the at least one processor is configured to use a trained machine learning model to predict the at least one class for the image patch.

14. The electronic device of claim 8 , wherein:
to filter each image patch, the at least one processor is configured to determine one or more dominant edges in the image patch; and
the one or more transform coefficients of the image patch are suppressed while preserving one or more other spectral coefficients relevant to the one or more dominant edges.

15. A non-transitory machine-readable medium containing instructions that when executed cause at least one processor of an electronic device to:
divide an image into overlapping image patches each having a specified size;
analyze content of each image patch using a mathematical transform technique to classify each image patch into at least one class, wherein the at least one class includes at least one of: a dark patch, a bright patch, or a patch having strong edge content;
filter each image patch for noise suppression by suppressing one or more transform coefficients of the image patch, wherein an amount of suppression for each of the one or more transform coefficients is selected according to the at least one class of the image patch such that (i) the amount of suppression for image patches classified as dark patches is increased and (ii) the amount of suppression for image patches classified as having strong edge content is reduced; and
reconstruct the filtered image patches into an output image.

16. The non-transitory machine-readable medium of claim 15 , wherein the instructions when executed further cause the at least one processor, before dividing the image into the overlapping image patches, to:
perform a gamma correction on the image; and
convert the image from a red-green-blue (RGB) domain to a luma-chroma (YUV) domain.

17. The non-transitory machine-readable medium of claim 16 , wherein the instructions that when executed cause the at least one processor to analyze the content of each image patch comprise instructions that when executed cause the at least one processor to:
determine multiple transform coefficients of the image patch, the multiple transform coefficients comprising the one or more suppressed transform coefficients;
determine multiple luma values of the image patch based on the multiple transform coefficients of the image patch; and
classify the image patch as a dark pa

Assignees

• Samsung Electronics Co Ltd

Inventors

• Nadir Zeeshan
• Sheikh Hamid R
• Glotzbach John W

Classifications

• G06T5/70 Primary

  Denoising; Smoothing · CPC title

• G06V10/56

  relating to colour · CPC title

• G06T2207/10024

  Color image · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06V10/60

  relating to illumination properties, e.g. using a reflectance or lighting model · CPC title