Granted Patent B2

US 12,198,407 · App. 17/710,254 · Granted Jan 14, 2025

Inventors: Han Zheng (Shenzhen, CN); Hong Shang (Shenzhen, CN); Xiaoning Wang (Shenzhen, CN); Jianhua Yao (Shenzhen, CN)

Assignee: TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED

G06V10/774G06V10/761G06V10/764G06V10/7715G06V10/776G06V10/82G06V2201/03

Abstract

Claims (70)

1. An artificial intelligence-based image processing method, comprising:

obtaining a first sample image of a source domain and a second sample image of a target domain, the first sample image of the source domain carrying a corresponding target processing result, and the first sample image and the second sample image being sample human tissue images;

encoding, by an encoding network of an image conversion network, the first sample image according to an image distribution difference between a target encoding space and the first sample image, to obtain a first sample image code;

generating, by a generation network of the image conversion network, based on the first sample image code and according to an image distribution difference between the second sample image and the target encoding space, a target sample image that conforms to an image distribution rule of the second sample image, the target sample image carrying a corresponding target processing result;

training a first image processing model based on the target sample image and the target processing result corresponding to the target sample image, to obtain a second image processing model; and

inputting, in response to obtaining a human tissue image of the target domain, the human tissue image into the second image processing model, positioning, by the second image processing model, a target human tissue in the human tissue image, and outputting position information of the target human tissue in the human tissue image.

2. The method according to claim 1 , wherein training the image conversion network comprises:

inputting the first sample image and the second sample image into an initial image conversion network, and encoding, by the initial image conversion network, the first sample image according to the image distribution difference of the target encoding space and the first sample image on the image distribution rule, to obtain the first sample image code;

encoding the second sample image according to the image distribution difference of the target encoding space and the second sample image on the image distribution rule, to obtain a second sample image code;

generating, according to the image distribution difference of the first sample image and the target encoding space on the image distribution rule, a third sample image and a fourth sample image based on the first sample image code and the second sample image code;

generating, according to the image distribution difference of the second sample image and the target encoding space on the image distribution rule, a fifth sample image and a sixth sample image based on the first sample image code and the second sample image code;

obtaining an image classification result of the first sample image, the second sample image, the third sample image, the fourth sample image, the fifth sample image, and the sixth sample image, the image classification result being used for indicating that an image is an original image or a generated image; and

adjusting a network parameter of the initial image conversion network according to an accuracy rate of the image classification result, to obtain the image conversion network.

3. The method according to claim 2 , further comprising:

respectively obtaining a first similarity between the third sample image and the fourth sample image, a second similarity between the fifth sample image and the sixth sample image, a third similarity between the first sample image and the third sample image, a fourth similarity between the first sample image and the fifth sample image, a fifth similarity between the second sample image and the fourth sample image, and a sixth similarity between the second sample image and the sixth sample image; and

adjusting the network parameter comprises:

adjusting the network parameter of the initial image conversion network according to the accuracy rate of the image classification result, the first similarity, the second similarity, the third similarity, the fourth similarity, the fifth similarity, and the sixth similarity.

4. The method according to claim 3 , further comprising:

inputting the fourth sample image and the sixth sample image into the initial image conversion network, and encoding, by the initial image conversion network, the fourth sample image and the sixth sample image, to obtain a third sample image code and a fourth sample image code; and

respectively obtaining a seventh similarity between the third sample image code and the first sample image code and an eighth similarity between the fourth sample image code and the second sample image code; and

adjusting the network parameter further comprises:

adjusting the network parameter of the initial image conversion network according to the seventh similarity and the eighth similarity.

5. The method according to claim 4 , wherein the first target condition is that a difference between the accuracy rate of the image classification result and a target accuracy rate is less than a difference threshold; or

the first target condition is that the difference between the accuracy rate of the image classification result and the target accuracy rate is less than the difference threshold, and the first similarity, the second similarity, the third similarity, the fourth similarity, the fifth similarity, and the sixth similarity converge; or

the first target condition is that the difference between the accuracy rate of the image classification result and the target accuracy rate is less than the difference threshold, and the first similarity, the second similarity, the third similarity, the fourth similarity, the fifth similarity, the sixth similarity, the seventh similarity, and the eighth similarity converge; or

the first target condition is that a quantity of iterations reaches a target quantity.

6. The method according to claim 1 , wherein training the first image processing model comprises:

inputting the target sample image into the first image processing model, positioning, by using the first image processing model, a target human tissue in the target sample image, and outputting predicted position information of the target human tissue in the target sample image;

obtaining accuracy of the predicted position information based on the predicted position information and target position information of the target human tissue in the target processing result; and

adjusting a model parameter of the first image processing model based on the accuracy, to obtain the second image processing model.

7. The method according to claim 1 , wherein positioning the target human tissue in the human tissue image comprises:

performing, by the second image processing model, feature extraction on the human tissue image, to obtain an image feature of the human tissue image;

obtaining a feature map of the human tissue image according to the image feature, each pixel in the feature map being used for indicating whether one region of the human tissue image comprises the target human tissue; and

obtaining and outputting the position information of the target human tissue according to the feature map and a target offset.

8. An artificial intelligence-based image processing apparatus, comprising: a memory storing computer program instructions; and a processor coupled to the memory and configured to execute the computer program instructions and perform:

obtaining a first sample image of a source domain and a second sample image of a target domain, the first sample image of the source domain carrying a corresponding target processing result, and the first sample image and the second sample image being sample human tissue images;

encoding, by an encoding network of an image conversion network, the first sample image according to an image distribution difference between a target encoding space and the first sample image, to obtain a first sample image code;

generating, by a generation network of the image conversion network, based on the first sample image code and according to an image distribution difference between the second sample image and the target encoding space, a target sample image that conforms to an image distribution rule of the second sample image, the target sample image carrying a corresponding target processing result;

training a first image processing model based on the target sample image and the target processing result corresponding to the target sample image, to obtain a second image processing model; and

inputting, in response to obtaining a human tissue image of the target domain, the human tissue image into the second image processing model, positioning, by the second image processing model, a target human tissue in the human tissue image, and outputting position information of the target human tissue in the human tissue image.

9. The apparatus of claim 8 , wherein training the image conversion network includes:

inputting the first sample image and the second sample image into an initial image conversion network, and encoding, by the initial image conversion network, the first sample image according to the image distribution difference of the target encoding space and the first sample image on the image distribution rule, to obtain the first sample image code;

encoding the second sample image according to the image distribution difference of the target encoding space and the second sample image on the image distribution rule, to obtain a second sample image code;

generating, according to the image distribution difference of the first sample image and the target encoding space on the image distribution rule, a third sample image and a fourth sample image based on the first sample image code and the second sample image code;

generating, according to the image distribution difference of the second sample image and the target encoding space on the image distribution rule, a fifth sample image and a sixth sample image based on the first sample image code and the second sample image code;

obtaining an image classification result of the first sample image, the second sample image, the third sample image, the fourth sample image, the fifth sample image, and the sixth sample image, the image classification result being used for indicating that an image is an original image or a generated image; and

adjusting a network parameter of the initial image conversion network according to an accuracy rate of the image classification result, to obtain the image conversion network.

10. The apparatus of claim 9 , wherein the processor is further configured to execute the computer program instructions and perform:

respectively obtaining a first similarity between the third sample image and the fourth sample image, a second similarity between the fifth sample image and the sixth sample image, a third similarity between the first sample image and the third sample image, a fourth similarity between the first sample image and the fifth sample image, a fifth similarity between the second sample image and the fourth sample image, and a sixth similarity between the second sample image and the sixth sample image; and

adjusting the network parameter comprises:

adjusting the network parameter of the initial image conversion network according to the accuracy rate of the image classification result, the first similarity, the second similarity, the third similarity, the fourth similarity, the fifth similarity, and the sixth similarity.

11. The apparatus of claim 10 , wherein the processor is further configured to execute the computer program instructions and perform:

inputting the fourth sample image and the sixth sample image into the initial image conversion network, and encoding, by the initial image conversion network, the fourth sample image and the sixth sample image, to obtain a third sample image code and a fourth sample image code; and

respectively obtaining a seventh similarity between the third sample image code and the first sample image code and an eighth similarity between the fourth sample image code and the second sample image code; and

adjusting the network parameter further comprises:

adjusting the network parameter of the initial image conversion network according to the seventh similarity and the eighth similarity.

12. The apparatus of claim 11 , wherein the first target condition is that a difference between the accuracy rate of the image classification result and a target accuracy rate is less than a difference threshold; or

the first target condition is that the difference between the accuracy rate of the image classification result and the target accuracy rate is less than the difference threshold, and the first similarity, the second similarity, the third similarity, the fourth similarity, the fifth similarity, and the sixth similarity converge; or

the first target condition is that the difference between the accuracy rate of the image classification result and the target accuracy rate is less than the difference threshold, and the first similarity, the second similarity, the third similarity, the fourth similarity, the fifth similarity, the sixth similarity, the seventh similarity, and the eighth similarity converge; or

the first target condition is that a quantity of iterations reaches a target quantity.

13. The apparatus of claim 8 , wherein training the first image processing model includes:

inputting the target sample image into the first image processing model, positioning, by using the first image processing model, a target human tissue in the target sample image, and outputting predicted position information of the target human tissue in the target sample image;

obtaining accuracy of the predicted position information based on the predicted position information and target position information of the target human tissue in the target processing result; and

adjusting a model parameter of the first image processing model based on the accuracy, to obtain the second image processing model.

14. A non-transitory computer-readable storage medium storing computer program instructions executable by at least one processor to perform:

obtaining a first sample image of a source domain and a second sample image of a target domain, the first sample image of the source domain carrying a corresponding target processing result, and the first sample image and the second sample image being sample human tissue images;

encoding, by an encoding network of an image conversion network, the first sample image according to an image distribution difference between a target encoding space and the first sample image, to obtain a first sample image code;

generating, by a generation network of the image conversion network, based on the first sample image code and according to an image distribution difference between the second sample image and the target encoding space, a target sample image that conforms to an image distribution rule of the second sample image, the target sample image carrying a corresponding target processing result;

training a first image processing model based on the target sample image and the target processing result corresponding to the target sample image, to obtain a second image processing model; and

inputting, in response to obtaining a human tissue image of the target domain, the human tissue image into the second image processing model, positioning, by the second image processing model, a target human tissue in the human tissue image, and outputting position information of the target human tissue in the human tissue image.

Assignments (1)

Priority Claims (1)

Continuity (2)

References Cited (20)