Image processing method and apparatus

Patent metadata

Field	Value
Publication number	US-12579798-B2
Application number	US-202218064132-A
Country	US
Kind code	B2
Filing date	Dec 9, 2022
Priority date	Jun 12, 2020
Publication date	Mar 17, 2026
Grant date	Mar 17, 2026

Abstract

An image processing method and apparatus are provided. The method includes: after image data of a target image is received, performing processing on the image data based on a network parameter to obtain enhanced image feature data of the target image; and performing processing on the target image based on the enhanced image feature data. The target image is a low-quality image, and the network parameter is used to indicate a correspondence between feature data of a low-quality image and feature data of a clear image. According to the application, a processing effect of a low-quality image is approved.

First claim

What is claimed is:

1 . An image processing method applied to an image processing apparatus, the method comprising:
receiving image data of a target image, wherein the target image is a low-quality image; obtaining feature data of the target image based on the image data; performing center-surround convolution computing on the feature data and the image data based on a network parameter to obtain enhanced image feature data of the target image, wherein the center-surround convolution computing simulates a light sensation principle of bipolar cells in a retina, and wherein the network parameter is used to indicate a correspondence between feature data of a low-quality image and feature data of a clear image; and performing processing on the target image based on the enhanced image feature data.

2 . The image processing method according to claim 1 ,
wherein the feature data is obtained by performing computing on the image data by using N layers of neural networks, and N is an integer greater than 0 and less than a preset threshold; wherein the performing the center-surround convolution computing on the feature data and the image data based on the network parameter to obtain the enhanced image feature data of the target image comprises: performing neural network computing on the feature data and the image data based on the network parameter to obtain residual data, wherein the residual data is used to indicate a deviation between the feature data of the target image and feature data of a clear image; and obtaining the enhanced image feature data of the target image based on the residual data and the feature data.

3 . The image processing method according to claim 2 , wherein the performing the neural network computing on the feature data and the image data based on the network parameter comprises:
performing at least first-level center-surround convolution computing, second-level center-surround convolution computing, and third-level center-surround convolution computing on the feature data and the image data based on the network parameter.

4 . The image processing method according to claim 3 , wherein input data of the first-level center-surround convolution computing comprises: the feature data and the image data,
wherein input data of the second-level center-surround convolution computing comprises: a computing result of the first-level center-surround convolution computing, and wherein input data of the third-level center-surround convolution computing comprises: a computing result of the second-level center-surround convolution computing.

5 . The image processing method according to claim 3 , wherein the residual data is obtained based on a computing result of the first-level center-surround convolution computing, a computing result of the second-level center-surround convolution computing, and a computing result of the third-level center-surround convolution computing.

6 . The image processing method according to claim 3 , wherein the first-level center-surround convolution computing is used to simulate a response of a central region in a retina of a human eye to the target image,
wherein the second-level center-surround convolution computing is used to simulate a response of a surrounded region of the retina of the human eye to the target image, and wherein the third-level center-surround convolution computing is used to simulate a response of a marginal region of the retina of the human eye to the target image.

7 . The image processing method according to claim 3 , wherein the first-level center-surround convolution computing comprises:
performing a first convolution operation on the feature data and the image data based on a first convolution kernel to obtain a first intermediate result, wherein a central-region weight of the first convolution kernel is 0; performing a second convolution operation on the feature data and the image data based on a second convolution kernel to obtain a second intermediate result, wherein the second convolution kernel comprises only a central-region weight, and the first convolution kernel and the second convolution kernel have a same size; and obtaining the computing result of the first-level center-surround convolution computing based on the first intermediate result and the second intermediate result.

8 . The image processing method according to claim 3 , wherein the second-level center-surround convolution computing comprises:
performing a third convolution operation on the computing result of the first-level center-surround convolution computing based on a third convolution kernel to obtain a third intermediate result, wherein a central-region weight of the third convolution kernel is 0; performing a fourth convolution operation on the computing result of the first-level center-surround convolution computing based on a fourth convolution kernel to obtain a fourth intermediate result, wherein the fourth convolution kernel comprises only a central-region weight, and the third convolution kernel and the fourth convolution kernel have a same size; and obtaining the computing result of the second-level center-surround convolution computing based on the third intermediate result and the fourth intermediate result.

9 . The image processing method according to claim 3 , wherein the third-level center-surround convolution computing comprises:
performing a fifth convolution operation on the computing result of the second-level center-surround convolution computing based on a fifth convolution kernel to obtain a fifth intermediate result, wherein a central-region weight of the fifth convolution kernel is 0; performing a sixth convolution operation on the computing result of the second-level center-surround convolution computing based on a sixth convolution kernel to obtain a sixth intermediate result, wherein the sixth convolution kernel comprises only a central-region weight, and the fifth convolution kernel and the sixth convolution kernel have a same size; and obtaining the computing result of the third-level center-surround convolution computing based on the fifth intermediate result and the sixth intermediate result.

10 . The image processing method according to claim 1 , wherein the image processing apparatus is a neural network device, and the network parameter is obtained by training.

11 . An image processing apparatus comprising:
a memory comprising processor-executable instructions; and a processor in communication with the memory, wherein the processor is configured to execute the processor-executable instructions to facilitate the image processing apparatus to: receive image data of a target image, wherein the target image is a low-quality image; obtain feature data of the target image based on the image data; perform center-surround convolution computing on the feature data and the image data based on a network parameter to obtain enhanced image feature data of the target image, wherein the center-surround convolution computing simulates a light sensation principle of bipolar cells in a retina, and wherein the network parameter is used to indicate a correspondence between feature data of a low-quality image and feature data of a clear image; and perform processing on the target image based on the enhanced image feature data.

12 . The image processing apparatus according to claim 11 ,
wherein the feature data is obtained by performing computing on the image data by using N layers of neural networks, and N is an integer greater than 0 and less than a preset threshold; wherein the processor is further configured to execute the processor-executable instructions to facilitate the

Assignees

• Huawei Tech Co Ltd
• Univ Science & Technology China

Inventors

• Cheng Jie
• Jiang Lei
• Cao Yang
• Zha Zhengjun

Classifications

• G06T5/60

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

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T5/20

  using local operators · CPC title

• G06V10/774

  Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title