Alpha channel prediction

Patent metadata

Field	Value
Publication number	US-2022014773-A1
Application number	US-202117354562-A
Country	US
Kind code	A1
Filing date	Jun 22, 2021
Priority date	Jul 7, 2020
Publication date	Jan 13, 2022
Grant date	—

Abstract

Image coding using alpha channel prediction may include generating a reconstructed image using alpha channel prediction and outputting the reconstructed image. Generating the reconstructed image using alpha channel prediction may include decoding reconstructed color channel values for a current pixel expressed with reference to first color space, obtaining color space converted color channel values for the current pixel by converting the reconstructed color channel values to a second color space, obtaining an alpha channel lower bound for an alpha channel value for the current pixel using the color space converted color channel values, generating a candidate predicted alpha value for the current pixel, obtaining an adjusted predicted alpha value for the current pixel using the candidate predicted alpha value and the alpha channel lower bound, generating a reconstructed pixel for the current pixel using the adjusted predicted alpha value, and including the reconstructed pixel in the reconstructed image.

First claim

What is claimed is:

1 . A method, comprising:
generating a reconstructed image, wherein generating the reconstructed image includes generating the reconstructed image using alpha channel prediction, wherein generating the reconstructed image using alpha channel prediction includes:
obtaining reconstructed color channel values for a current pixel of the current image expressed with reference to first color space;
obtaining color space converted color channel values for the current pixel by converting the reconstructed color channel values to a second color space;
obtaining an alpha channel lower bound for an alpha channel value for the current pixel using the color space converted color channel values;
generating a candidate predicted alpha value for the current pixel;
obtaining an adjusted predicted alpha value for the current pixel using the candidate predicted alpha value and the alpha channel lower bound;
generating a reconstructed pixel for the current pixel using the adjusted predicted alpha value; and
including the reconstructed pixel in the reconstructed image; and
outputting the reconstructed image.

2 . The method of claim 1 , wherein:
the first color space is the YUV color space; the reconstructed color channel values include a luminance channel value, a first chrominance channel value, and a second chrominance channel value; and obtaining the reconstructed color channel values includes decoding residual color channel values from an encoded bitstream.

3 . The method of claim 1 , wherein the second color space is the RGB color space and the color space converted color channel values include a red color channel value, a green color channel value, and a blue color channel value.

4 . The method of claim 3 , wherein obtaining the alpha channel lower bound includes:
obtaining a normalized red color channel value by dividing the red color channel value by a defined maximum value for the red color channel; obtaining a normalized green color channel value by dividing the green color channel value by a defined maximum value for the green color channel; obtaining a normalized blue color channel value by dividing the blue color channel value by a defined maximum value for the blue color channel; identifying a maximum value among the normalized red color channel value, the normalized green color channel value, and the normalized blue color channel value; and identifying, as the alpha channel lower bound, a product of multiplying the maximum value by a defined maximum value for the alpha channel.

5 . The method of claim 1 , wherein generating the candidate predicted alpha value includes:
identifying a previously reconstructed context pixel for predicting the candidate predicted alpha value; and obtaining the candidate predicted alpha value using the previously reconstructed context pixel.

6 . The method of claim 1 , wherein obtaining the adjusted predicted alpha value includes:
identifying, as the adjusted predicted alpha value, a maximum value among the candidate predicted alpha value and the alpha channel lower bound.

7 . A method, comprising:
generating an encoded image, wherein generating the encoded image includes generating the encoded image using alpha channel prediction, wherein generating the encoded image using alpha channel prediction includes:
identifying a current pixel from an input image, wherein the current pixel includes input color channel values, wherein the input color channel values are expressed with reference to first color space, and wherein the input color channel values include an input alpha channel value;
obtaining pre-multiplied color channel values for the pixel using the input color channel values;
obtaining reconstructed color values for the pixel using the pre-multiplied color channel values, wherein the reconstructed color channel values are expressed with reference to second color space;
obtaining color space converted color channel values for the current pixel by color space converting the reconstructed color channel values to the first color space;
obtaining an alpha channel lower bound for a reconstructed alpha channel value for the current pixel using the color space converted color channel values;
generating a candidate predicted alpha value for the current pixel;
obtaining an adjusted predicted alpha value for the current pixel using the candidate predicted alpha value and the alpha channel lower bound;
obtaining a residual alpha value as a difference of subtracting the adjusted predicted alpha value from the input alpha channel value; and
including the residual alpha value in an output bitstream; and
outputting the output bitstream.

8 . The method of claim 7 , wherein the second color space is the YUV color space and the reconstructed color channel values include a luminance channel value, a first chrominance channel value, and a second chrominance channel value.

9 . The method of claim 7 , wherein:
the first color space is the RGB color space; the input color channel values include an input red color channel value, an input green color channel value, and an input blue color channel value; the pre-multiplied color channel values include a pre-multiplied red color channel value, a pre-multiplied green color channel value, and a pre-multiplied blue color channel value; and the color space converted color channel values include a color space converted red color channel value, a color space converted green color channel value, and a color space converted blue color channel value.

10 . The method of claim 9 , wherein obtaining the pre-multiplied color channel values includes:
obtaining, as the pre-multiplied red color channel value, a product of multiplying the input red color channel value by the input alpha channel value; obtaining, as the pre-multiplied green color channel value, a product of multiplying the input green color channel value by the input alpha channel value; and obtaining, as the pre-multiplied blue color channel value, a product of multiplying the input blue color channel value by the input alpha channel value.

11 . The method of claim 9 , wherein obtaining the alpha channel lower bound includes:
obtaining a normalized red color channel value by dividing the color space converted red color channel value by a defined maximum value for the red color channel; obtaining a normalized green color channel value by dividing the color space converted green color channel value by a defined maximum value for the green color channel; obtaining a normalized blue color channel value by dividing the color space converted blue color channel value by a defined maximum value for the blue color channel; identifying a maximum value among the normalized red color channel value, the normalized green color channel value, and the normalized blue color channel value; and identifying, as the alpha channel lower bound, a product of multiplying the maximum value by a defined maximum value for the alpha channel.

12 . The method of claim 7 , wherein obtaining the reconstructed color values includes:
color space converting the pre-multiplied color channel values to the second color space; obtaining respective predicted color channel values for the pre-multiplied color channel values; obtaining respective residual color channel values as respective differences of subtracting the predicted color channel values from the corresponding pre-multiplied color channel values; lossily encoding the residual color channel values to obtain encoded residual color channel values; and obtaining, as the reconstructed color values, respective sums of adding the encoded residual c

Assignees

• Google Llc

Inventors

• Ustarroz-Calonge Maryla
• Rabaud Vincent

Classifications

• H04N19/91

  Entropy coding, e.g. variable length coding [VLC] or arithmetic coding · CPC title

• H04N19/61

  in combination with predictive coding · CPC title

• H04N19/147

  according to rate distortion criteria (rate-distortion as a criterion for motion estimation H04N19/567) · CPC title

• H04N19/176 Primary

  the region being a block, e.g. a macroblock · CPC title

• H04N19/96

  Tree coding, e.g. quad-tree coding · CPC title