Method and apparatus for processing image

Patent metadata

Field	Value
Publication number	US-11734809-B2
Application number	US-202117174002-A
Country	US
Kind code	B2
Filing date	Feb 11, 2021
Priority date	Apr 23, 2020
Publication date	Aug 22, 2023
Grant date	Aug 22, 2023

Abstract

Embodiments of the present disclosure provide a method and apparatus for processing an image, and relates to the field of computer vision technology. The method may include: acquiring a value to be processed, where the value to be processed is associated with an image to be processed; and processing the value to be processed by using a quality scoring model to generate a score of the image to be processed in a target scoring domain, where the score of the image to be processed in the target scoring domain is related to an image quality of the image to be processed.

First claim

What is claimed is:

1. A method for processing an image, the method comprising:
acquiring a value to be processed, wherein the value to be processed is associated with an image to be processed; and
processing the value to be processed by using a quality scoring model to generate a score of the image to be processed in a target scoring domain, wherein the score of the image to be processed in the target scoring domain is related to an image quality of the image to be processed;
wherein the value to be processed is a score of the image to be processed in an original scoring domain; and
processing the value to be processed by using the quality scoring model to generate the score of the image to be processed in the target scoring domain comprises:
inputting the score of the image to be processed in the original scoring domain into the quality scoring model to obtain the score of the image to be processed in the target scoring domain, wherein the quality scoring model is a monotonic neural network, and a number of hidden units in the monotonic neural network is smaller than a preset threshold.

2. The method according to claim 1 , wherein the quality scoring model comprises a scoring network and a monotonic neural network; and acquiring the value to be processed comprises:
inputting the image to be processed into the scoring network to obtain an initial score, output from the scoring network, of the image to be processed; and
processing the value to be processed by using the quality scoring model to generate the score of the image to be processed in the target scoring domain comprises:
inputting the initial score into the monotonic neural network to obtain the score of the image to be processed in the target scoring domain, wherein a number of hidden units in the monotonic neural network is smaller than a preset threshold.

3. The method according to claim 2 , wherein the quality scoring model comprises at least two monotonic neural networks, and different monotonic neural networks in the at least two monotonic neural networks correspond to different scoring domains; and
inputting the initial score into the monotonic neural network to obtain the score of the image to be processed in the target scoring domain comprises:
inputting the initial score into the at least two monotonic neural networks to obtain a score, output from each of the at least two monotonic neural networks, of the image to be processed in a scoring domain corresponding to the monotonic neural network.

4. The method according to claim 2 , wherein the method further comprises:
acquiring a training sample set, wherein a training sample in the training sample set comprises a sample image and a reference score of the sample image in a specified scoring domain, and the specified scoring domain and the target scoring domain are different scoring domains;
inputting the sample image into the scoring network to obtain an initial score of the sample image;
inputting the initial score of the sample image into a monotonic neural network to be trained to obtain a predicted score of the sample image in the specified scoring domain; and
determining a loss value of the predicted score based on the reference score and the predicted score, and training the monotonic neural network to be trained by means of the loss value to obtain a trained monotonic neural network.

5. An electronic device, comprising:
one or more processors; and
a memory for storing one or more programs,
wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform operations, the operations comprising:
acquiring a value to be processed, wherein the value to be processed is associated with an image to be processed; and
processing the value to be processed by using a quality scoring model to generate a score of the image to be processed in a target scoring domain, wherein the score of the image to be processed in the target scoring domain is related to an image quality of the image to be processed;

wherein the value to be processed is a score of the image to be processed in an original scoring domain; and
processing the value to be processed by using the quality scoring model to generate the score of the image to be processed in the target scoring domain comprises:
inputting the score of the image to be processed in the original scoring domain into the quality scoring model to obtain the score of the image to be processed in the target scoring domain wherein the quality scoring model is a monotonic neural network and a number of hidden units in the monotonic neural network is smaller than a preset threshold.

6. The electronic device according to claim 5 , wherein the quality scoring model comprises a scoring network and a monotonic neural network; and acquiring the value to be processed comprises:
inputting the image to be processed into the scoring network to obtain an initial score, output from the scoring network, of the image to be processed; and
processing the value to be processed by using the quality scoring model to generate the score of the image to be processed in the target scoring domain comprises:
inputting the initial score into the monotonic neural network to obtain the score of the image to be processed in the target scoring domain, wherein a number of hidden units in the monotonic neural network is smaller than a preset threshold.

7. The electronic device according to claim 6 , wherein the quality scoring model comprises at least two monotonic neural networks, and different monotonic neural networks in the at least two monotonic neural networks correspond to different scoring domains; and
inputting the initial score into the monotonic neural network to obtain the score of the image to be processed in the target scoring domain comprises:
inputting the initial score into the at least two monotonic neural networks to obtain a score, output from each of the at least two monotonic neural networks, of the image to be processed in a scoring domain corresponding to the monotonic neural network.

8. The electronic device according to claim 6 , wherein the operations further comprise:
acquiring a training sample set, wherein a training sample in the training sample set comprises a sample image and a reference score of the sample image in a specified scoring domain, and the specified scoring domain and the target scoring domain are different scoring domains;
inputting the sample image into the scoring network to obtain an initial score of the sample image;
inputting the initial score of the sample image into a monotonic neural network to be trained to obtain a predicted score of the sample image in the specified scoring domain; and
determining a loss value of the predicted score based on the reference score and the predicted score, and training the monotonic neural network to be trained by means of the loss value to obtain a trained monotonic neural network.

9. A non-transitory computer-readable storage medium, storing a computer program thereon, wherein the computer program, when executed by a processor, causes the processor to perform operations, the operations comprising:
acquiring a value to be processed, wherein the value to be processed is associated with an image to be processed; and
processing the value to be processed by using a quality scoring model to generate a score of the image to be processed in a target scoring domain, wherein the score of the image to be processed in the target scoring domain is related to an image quality of the image to be processed;
wherein the value to be processed is a score of the image to be processed in an original scoring domain; and
processing the value to be processed by u

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Long Xiang
• Wang Ping
• Zhou Zhichao
• Li Fu
• He Dongliang
• Sun Hao

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06T7/0002 Primary

  Inspection of images, e.g. flaw detection · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• G06N3/08

  Learning methods · CPC title