Granted Patent B2

US 12,657,875 · App. 18/565,030 · Granted Jun 16, 2026

Inventors: Tong Liu (Beijing, CN); Zeyu Shangguan (Beijing, CN)

Assignee: BOE Technology Group Co., Ltd.

G06V10/764G06T7/11G06V10/44G06T2207/20081G06T2207/20084G06T2207/20132

Abstract

Claims (65)

1 . A method for training an image processing model, applicable to an image processing device, and the method comprising:

acquiring a plurality of original image sets, wherein each of the original image sets comprises a plurality of original images of a same category, and categories of the original images comprised in different original image sets are different;

acquiring a training sample set by cropping the plurality of original images in the plurality of original image sets, wherein the training sample set comprises a plurality of training samples, and each of the training samples represents a subimage acquired by cropping any of the plurality of original images;

determining a plurality of positive sample pairs and a plurality of negative sample pairs from the training sample set, wherein each of the positive sample pairs comprises two training samples acquired based on different original images in the same original image set, and each of the negative sample pairs comprises two training samples acquired based on the original images in different original image sets; and

training the image processing model based on the plurality of positive sample pairs and the plurality of negative sample pairs;

wherein determining the plurality of positive sample pairs from the training sample set comprises:

determining a plurality of candidate sample pairs from the training sample set, wherein each of the candidate sample pairs comprises two training samples acquired based on different original images in the same original image set;

for each of the candidate sample pairs, extracting a feature vector of each training sample in the candidate sample pair based on a convolutional neural network;

acquiring the similarity of the candidate sample pair by processing feature vectors of two training samples in the candidate sample pair based on a similarity measurement algorithm; and

determining the candidate sample pair with the similarity greater than a similarity threshold as the positive sample pair.

2 . The method for training an image processing model according to claim 1 , wherein cropping the plurality of original image samples in the plurality of original image sets comprises:

generating a cropping size within a target size range randomly for each original image to be cropped in the plurality of original image sets;

determining a reference point of a cropping area based on a size of the original image and the cropping size; and

determining the cropping area in the original image based on the cropping size and the reference point, and cropping the cropping area.

3 . The method for training an image processing model according to claim 2 , wherein the target size range comprises a width range and a height range, and the cropping size comprises a width within the width range and a height within the height range; and

the cropping area is a rectangular area, and the reference point of the cropping area is a vertex of the rectangular area or a central point of the rectangular area.

4 . The method for training an image processing model according to claim 1 , wherein determining the plurality of negative sample pairs from the training sample sets comprises:

determining a plurality of negative sample pairs from the training sample set, wherein a number of the negative sample pairs is equal to a number of the plurality of positive sample pairs.

5 . The method for training an image processing model according to claim 1 , wherein training the image processing model based on the plurality of positive sample pairs and the plurality of negative sample pairs comprises:

marking a ground truth of each of the positive sample pairs as 1 and a ground truth of each of the negative sample pairs as 0; and

training the image processing model based on the plurality of marked positive sample pairs and the plurality of marked negative sample pairs.

6 . An image classifying method, comprising:

acquiring a to-be-classified target image; and

inputting the target image into an image classifying model to acquire a category of the target image outputted by the image classifying model,

wherein the image classifying model is acquired by training based on the method for training an image processing model as defined in claim 1 .

7 . The image classifying method according to claim 6 , wherein inputting the target image into the image classifying model to acquire the category of the target image outputted by the image classifying model comprises:

inputting the target image into the image classifying model to acquire a similarity between the target image and reference images of different categories outputted by the image classifying model; and

determining a category of a reference image having a highest similarity with respect to the target image in the reference images of different categories as the category of the target image.

8 . The image classifying method according to claim 6 , wherein inputting the target image into the image classifying model to acquire the category of the target image outputted by the image classifying model comprises:

inputting the target image into the image classifying model to acquire a similarity between the target image and image features of different categories outputted by the image classifying model; and

determining a category of an image feature having a highest similarity with respect to the target image in the image features of different categories as the category of the target image,

wherein the image features of each category are acquired by performing feature extraction on a plurality of training samples of the category.

9 . An image processing device, comprising a processor and a memory storing one or more instructions thereon, wherein the one or more instructions are loaded and executed by the processor to implement:

acquiring a plurality of original image sets, wherein each of the original image sets comprises a plurality of original images of a same category, and categories of the original images comprised in different original image sets are different;

acquiring a training sample set by cropping the plurality of original images in the plurality of original image sets, wherein the training sample set comprises a plurality of training samples, and each of the training samples represents subimage acquired by cropping any of the plurality of original images;

determining a plurality of positive sample pairs and a plurality of negative sample pairs from the training sample set, wherein each of the positive sample pairs comprises two training samples acquired based on different original images in the same original image set, and each of the negative sample pairs comprises two training samples acquired based on the original images in different original image sets; and

training the image processing model based on the plurality of positive sample pairs and the plurality of negative sample pairs;

wherein the one or more instructions are loaded and executed by the processor to implement:

determining a plurality of candidate sample pairs from the training sample set wherein each of the candidate sample pairs comprises two training samples acquired based on different original images in the same original image set;

for each of the candidate sample pairs, extracting a feature vector of each training sample in the candidate sample pair based on a convolutional neural network;

acquiring the similarity of the candidate sample pair by processing feature vectors of two training samples in the candidate sample pair based on a similarity measurement algorithm; and

determining the candidate sample pair with the similarity greater than a similarity threshold as the positive sample pair.

10 . A non-transitory computer-readable storage medium storing one or more instructions thereon, wherein the one or more instructions are loaded and executed by a processor to implement:

acquiring a plurality of original image sets, wherein each of the original image sets comprises a plurality of original images of a same category, and categories of the original images comprised in different original image sets are different;

acquiring a training sample set by cropping the plurality of original images in the plurality of original image sets, wherein the training sample set comprises a plurality of training samples, and each of the training samples represents a subimage acquired by cropping any of the plurality of original images;

determining a plurality of positive sample pairs and a plurality of negative sample pairs from the training sample set, wherein each of the positive sample pairs comprises two training samples acquired based on different original images in the same original image set, and each of the negative sample pairs comprises two training samples acquired based on the original images in different original image sets; and

training the image processing model based on the plurality of positive sample pairs and the plurality of negative sample pairs;

wherein the one or more instructions are loaded and executed by a processor to implement:

determining a plurality of candidate sample pairs from the training sample set, wherein each of the candidate sample pairs comprises two training samples acquired based on different original images in the same original image set;

for each of the candidate sample pairs, extracting a feature vector of each training sample in the candidate sample pair based on a convolutional neural network;

acquiring the similarity of the candidate sample pair by processing feature vectors of two training samples in the candidate sample pair based on a similarity measurement algorithm; and

determining the candidate sample pair with the similarity greater than a similarity threshold as the positive sample pair.

11 . The image processing device according to claim 9 , wherein the one or more instructions are loaded and executed by the processor to implement:

generating a cropping size within a target size range randomly for each original image to be cropped in the plurality of original image sets;

determining a reference point of a cropping area based on a size of the original image and the cropping size; and

determining the cropping area in the original image based on the cropping size and the reference point, and cropping the cropping area.

12 . The image processing device according to claim 11 , wherein the target size range comprises a width range and a height range, and the cropping size comprises a width within the width range and a height within the height range; and

the cropping area is a rectangular area, and the reference point of the cropping area is a vertex of the rectangular area or a central point of the rectangular area.

13 . The image processing device according to claim 9 , wherein the one or more instructions are loaded and executed by the processor to implement:

determining a plurality of negative sample pairs from the training sample set, wherein a number of the negative sample pairs is equal to a number of the plurality of positive sample pairs.

14 . The image processing device according to claim 9 , wherein the one or more instructions are loaded and executed by the processor to implement:

marking a ground truth of each of the positive sample pairs as 1 and a ground truth of each of the negative sample pairs as 0; and

training the image processing model based on the plurality of marked positive sample pairs and the plurality of marked negative sample pairs.

15 . An image processing device, comprising a processor and a memory storing one or more instructions thereon, wherein the one or more instructions are loaded and executed by the processor to implement the image classifying method as defined in claim 6 .

16 . A non-transitory computer-readable storage medium storing one or more instructions thereon, wherein the one or more instructions are loaded and executed by a processor to implement the image classifying method as defined in claim 6 .

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (20)