Granted Patent B2

US 12,726,652 · App. 18/616,927 · Granted Sep 1, 2026

Inventors: Cheolkon Jung (Dongguan, CN); Hui Lan (Dongguan, CN); Zhe Ji (Dongguan, CN)

Assignee: GUANGDONG OPPO MOBILE TELECOMMUNICATIONS CORP., LTD.

H04N19/59H04N19/597

Abstract

Claims (26)

1 . A method for decoding, comprising:

receiving, by a processor, a first image and a second image of a scene, wherein the first image is downsampled and is different from the second image, the first image is a depth image, and the second image is a color image;

obtaining, by the processor, a residual map associated with edges of the scene according to the second image and the downsampled first image, comprising:

deriving an intensity edge map from the color image;

estimating a depth edge feature map based on the depth image using a first machine learning model;

estimating an intensity edge feature map based on the intensity edge map using a second machine learning model;

estimating a color edge feature map based on the color image, the depth image, and the intensity edge map using a third machine learning model;

fusing the estimated depth edge feature map and the estimated intensity edge feature map to the estimated color edge feature map; and

estimating the residual map based on the fused color edge feature map using a fourth machine learning model;

upsampling, by the processor, the downsampled first image; and

reconstructing, by the processor, the first image based on the upsampled first image and the residual map.

2 . The method of claim 1 , wherein the first image and the second image are complementary to one another with respect to the scene.

3 . The method of claim 1 , wherein estimating the depth edge feature map comprises:

extracting a depth edge feature from the depth image using a first sub-model of the first machine learning model; and

upsampling the extracted depth edge feature using a second sub-model of the first machine learning model.

4 . The method of claim 3 , wherein estimating the intensity edge feature map comprises:

downsampling the intensity edge map;

extracting a downsampled intensity edge feature from the downsampled intensity edge map using a first sub-model of the second machine learning model; and

extracting an intensity edge feature from the intensity edge map using a second sub-model of the second machine learning model.

5 . The method of claim 4 , wherein estimating the color edge feature map comprises:

extracting a color edge feature from the color image using a first sub-model of the third machine learning model;

downsampling the extracted color edge feature using a second sub-model of the third machine learning model;

fusing the extracted depth edge feature and the downsampled intensity edge feature to the downsampled color edge feature; and

upsampling the fused color edge feature based on the upsampled depth edge feature using a third sub-model of the third machine learning model.

6 . The method of claim 5 , wherein sizes of the extracted depth edge feature, the downsampled intensity edge feature, and the downsampled color edge feature are the same.

7 . The method of claim 1 , wherein reconstructing the first image comprises fusing the residual map to the upsampled first image.

Assignments (1)

Continuity (2)

References Cited (22)