Learning based discrete cosine noise filter

Patent metadata

Field	Value
Publication number	US-12340487-B2
Application number	US-202217665294-A
Country	US
Kind code	B2
Filing date	Feb 4, 2022
Priority date	Feb 4, 2022
Publication date	Jun 24, 2025
Grant date	Jun 24, 2025

Abstract

A method includes filtering an input image through a discrete cosine transform based noise filter (DCT-NF). The DCT-NF converts the input image from an original space into a perceptual space, applying gamma correction. The DCT-NF separates luminance channels of the input image from chroma channels of the input image. The DCT-NF divides the input image into overlapping patches and computes DCT transform of the patches. Each patch is a different partial portion of the input image. The DCT-NF suppresses patches that include an input DCT coefficient within a threshold range. The DCT-NF applies an inverse discrete cosine transform (IDCT) to the suppressed patches and remaining overlapping patches that include an input DCT coefficient outside the threshold range. The DCT-NF re-combines luminance and chroma channels of the IDCT-transformed patches. The DCT-NF generates a DCT noise-filtered output image by re-converting the IDCT-transformed patches to the original space by applying inverse gamma correction.

First claim

What is claimed is:

1. A method comprising:
receiving an input image from an image capturing device;
filtering the received input image by applying a discrete cosine transform based noise filtering (DCT-NF) process, the DCT-NF process comprising:
converting the input image from an original space into a perceptual space by applying a gamma correction to generate a gamma-corrected image;
separating a luminance channel from chroma channels of the input image or the gamma-corrected image to generate a luminance image composed of the luminance channel and chrominance images composed of the chroma channels, respectively;
dividing the luminance image into overlapping patches, each patch being a different partial portion of the luminance image, wherein the overlapping patches include a first grid of patches that at least partially overlaps a second grid of patches;
computing a discrete cosine transform (DCT) of the overlapping patches;
among the overlapping patches, suppressing patches that include an input DCT coefficient within a threshold range;
applying an inverse discrete cosine transform (IDCT) to the suppressed patches and to remaining overlapping patches that include an input DCT coefficient outside of the threshold range to generate IDCT-transformed patches;
combining a luminance channel of the IDCT-transformed patches and the chrominance images to form a color image; and
generating a DCT noise-filtered output image by applying an inverse gamma correction to convert the color image to the original space; and

outputting the DCT noise-filtered output image;
wherein the second grid of patches covers a different portion of the luminance image than the first grid of patches such that (i) the first grid of patches covers part of the luminance image not covered by the second grid of patches and (ii) the second grid of patches includes an empty portion not covering any portion of the luminance image.

2. The method of claim 1 , wherein the DCT-NF process further comprises:
tuning threshold parameters and suppression parameters defining the threshold range based on a power spectral density obtained from a live image capture.

3. The method of claim 1 , wherein at least one of threshold parameters or suppression parameters are learned from training data using a machine learning technique.

4. The method of claim 1 , wherein:
separating the luminance channel from the chroma channels comprises converting, from an RGB domain to a YUV domain, the input image or the gamma-corrected image; and
combining the luminance channel of the IDCT-transformed patches and the chroma images comprises converting from the YUV domain to the RGB domain.

5. The method of claim 4 , wherein the DCT-NF process further comprises:
separately filtering a U channel and a V channel of the input image or the gamma-corrected image through a low computational filter.

6. The method of claim 4 , wherein the DCT-NF process further comprises:
classifying patches among the overlapping patches; and
suppressing the classified patches based on threshold parameters and suppression parameters that vary spatially based upon the classification.

7. The method of claim 1 , wherein:
the first grid of patches includes a first grid of M×M patches; and
the second grid of patches includes a second grid of M×M patches.

8. The method of claim 1 , wherein the threshold range is defined by a first slope for input DCT coefficient values within a first suppression range, and a second slope for input DCT coefficient values within a second suppression range.

9. The method of claim 1 , further comprising computing an average of patches for reconstruction within the color image.

10. An electronic device comprising:
a display; and
at least one processor configured to:
receive an input image from an image capturing device;
filter the received input image by applying a discrete cosine transform based noise filtering (DCT-NF) process, wherein the at least one processor is configured to perform the DCT-NF process by:
converting the input image from an original space into a perceptual space by applying a gamma correction to generate a gamma-corrected image;
separating a luminance channel from chroma channels of the input image or the gamma-corrected image to generate a luminance image composed of the luminance channel and chrominance images composed of the chroma channels, respectively;
dividing the luminance image into overlapping patches, each patch being a different partial portion of the luminance image, wherein the overlapping patches include a first grid of patches that at least partially overlaps a second grid of patches;
computing a discrete cosine transform (DCT) of the overlapping patches;
among the overlapping patches, suppressing patches that include an input DCT coefficient within a threshold range;
applying an inverse discrete cosine transform (IDCT) to the suppressed patches and to remaining overlapping patches that include an input DCT coefficient outside of the threshold range to generate IDCT-transformed patches;
combining a luminance channel of the IDCT-transformed patches and the chrominance images to form a color image; and
generating a DCT noise-filtered output image by applying an inverse gamma correction to convert the color image to the original space; and

output the DCT noise-filtered output image for displaying on the display;
wherein the second grid of patches covers a different portion of the luminance image than the first grid of patches such that (i) the first grid of patches covers part of the luminance image not covered by the second grid of patches and (ii) the second grid of patches includes an empty portion not covering any portion of the luminance image.

11. The electronic device of claim 10 , wherein the DCT-NF process further comprises:
tuning threshold parameters and suppression parameters defining the threshold range based on a power spectral density obtained from a live image capture.

12. The electronic device of claim 10 , wherein at least one of threshold parameters or suppression parameters are learned from training data using a machine learning technique.

13. The electronic device of claim 10 , wherein the at least one processor is configured to:
separate the luminance channel from the chroma channels by converting, from an RGB domain to a YUV domain, the input image or the gamma-corrected image; and
combine the luminance channel of the IDCT-transformed patches and the chroma images by converting from the YUV domain to the RGB domain.

14. The electronic device of claim 13 , wherein the DCT-NF process further comprises:
separately filtering a U channel and a V channel of the input image or the gamma-corrected image through a low computational filter.

15. The electronic device of claim 13 , wherein the DCT-NF process further comprises:
classifying patches among the overlapping patches; and
suppressing the classified patches based on threshold parameters and suppression parameters that vary spatially based upon the classification.

16. The electronic device of claim 10 , wherein:
the first grid of patches includes a first grid of M×M patches; and
the second grid of patches includes a second grid of M×M patches.

17. The electronic device of claim 10 , wherein the threshold range is defined by a first slope for input DCT coefficient values within a first suppression range, and a second slope for input DCT coefficient values within a second suppression range.

18. The electronic device of claim 10 , wherein the at least one process

Assignees

• Samsung Electronics Co Ltd

Inventors

• Nadir Zeeshan
• Glotzbach John W
• Sheikh Hamid R

Classifications

• G06T5/10

  using non-spatial domain filtering · CPC title

• G06T2207/20056

  Discrete and fast Fourier transform, [DFT, FFT] · CPC title

• G06T5/60

  using machine learning, e.g. neural networks · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T5/70 Primary

  Denoising; Smoothing · CPC title