# US 2023/0069654 A1 · Method and device for processing image, method for training image processing model, and storage medium

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2023/0069654 A1
- Document ID: US20230069654A1
- Status: Published application
- Kind: A1
- Application number: 17/774,610
- Filed: 2021-05-14
- Published: 2023-03-02
- Earliest claimed: 2020-05-18
- Assignees: Boe Technology Group Co., Ltd.
- Applicants: BOE Technology Group Co., Ltd.
- Inventors: Hanwen LIU; Tingting WANG; Guannan CHEN
- Web page: https://patentgrep.com/patent/US20230069654A1

## CPC

- G06T 5/70: Denoising; Smoothing
- G06N 20/00: Machine learning
- G06T 5/60: using machine learning, e.g. neural networks
- G06T 5/73: Deblurring; Sharpening
- G06T 2207/20081: Training; Learning
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

An image processing method and an image processing model training method and device. The image processing method comprises: obtaining a first original image to be processed (S101), and downsampling and blurring said first original image according to a configured downsampling processing algorithm and blur parameter to obtain a degraded image (S102). In the method, a new photo can be downsampled and blurred to obtain the degraded image similar to an old photo, and a good training sample is provided for the application of machine learning in the old photo restoration technology. An image processing model applied in the old photo restoration technology is trained on the basis of the training sample, so as to obtain the image processing model having stronger restoration capability, and when the old photo is restored by using the image processing model, the restoration efficiency is higher, and the restoration effect is better.

## Claims

### Claim 1

A method for processing an image, comprising:
acquiring a first original image to be processed; and
acquiring a degraded image by performing down-sampling processing and blurring processing on the first original image according to a configured down-sampling processing algorithm and a blurring parameter;
wherein the down-sampling processing algorithm comprises a plurality of sampling steps.

### Claim 2 (depends on 1)

The method according to claim 1, wherein the blurring processing is Gaussian blurring processing, and the blurring parameter comprises at least one of a Gaussian kernel size, a horizontal variance, and a vertical variance.

### Claim 3 (depends on 1)

The method according to claim 1, wherein the down-sampling processing algorithm comprises at least two of a bicubic interpolation algorithm-based sampling step, a bilinear interpolation algorithm-based sampling step, and a Lanczos algorithm-based sampling step.

### Claim 4 (depends on 1)

The method according to claim 1, wherein a sampling magnification of each of the sampling steps in the down-sampling processing algorithm is within a first magnification range, and an upper limit of the first magnification range is less than or equal to 1.

### Claim 5 (depends on 4)

The method according to claim 4, wherein in the down-sampling processing algorithm, a sampling magnification of a bicubic interpolation algorithm-based sampling step is 0.5, a sampling magnification of a bilinear interpolation algorithm-based sampling step is 0.5, and a sampling magnification of a Lanczos algorithm-based sampling step is 0.25.

### Claim 6 (depends on 1)

The method according to claim 1, further comprising:
performing noise processing on the first original image according to a configured noise parameter.

### Claim 7 (depends on 6)

The method according to claim 6, wherein the noise processing is Gaussian noise processing, and the noise parameter comprises at least one of a mean and a standard deviation.

### Claim 8 (depends on 1)

The method according to claim 1, further comprising:
performing zoom blurring processing on the first original image according to a configured zoom blurring processing algorithm;
wherein the zoom blurring processing algorithm comprises a plurality of sampling steps, and a sampling magnification of each of the sampling steps in the zoom blurring processing algorithm is within a second magnification range, and an upper limit of the second magnification range is greater than 1.

### Claim 9 (depends on 8)

The method according to claim 8, wherein the zoom blurring processing algorithm comprises at least two of a bicubic interpolation algorithm-based sampling step, a bilinear interpolation algorithm-based sampling step, and a Lanczos algorithm-based sampling step.

### Claim 10 (depends on 8)

The method according to claim 8, wherein in the zoom blurring processing algorithm, a sampling magnification of a bicubic interpolation algorithm-based sampling step is 0.5, a sampling magnification of a bilinear interpolation algorithm-based sampling step is 2, and a sampling magnification of a Lanczos algorithm-based sampling step is 0.25.

### Claim 11 (depends on 1)

The method according to claim 1 , further comprising:
performing compression noise processing on the first original image according to a configured compression noise parameter.

### Claim 12 (depends on 11)

The method according to claim 11, wherein the compression noise parameter comprises a coding quality coefficient; and
said performing compression noise processing on the first original image according to the configured compression noise parameter comprises:
performing image coding on pixels of the first original image according to the coding quality coefficient.

### Claim 13 (depends on 12)

The method according to claim 12, wherein the coding quality coefficient is determined by:
Q
jepg
=
normal
ave
q
,var
q
,
wherein Qjepg represents the coding quality coefficient, aveq represents a mean of values within a preset coefficient range, varq represents a variance of the values within the preset coefficient range, normal() represents a normal distribution function, and normal(aveq, varq) represents returning to one of a set of numbers that obeys the normal distribution of the mean aveq and the variance varq.

### Claim 14 (depends on 2)

The method according to claim 2, wherein said acquiring a degraded image by performing down-sampling processing and blurring processing on the first original image comprises:
acquiring a first degraded image by performing down-sampling processing and blurring processing on the first original image sequentially; and
the method further comprises:
acquiring a second degraded image by performing Gaussian noise processing on the first degraded image according to a configured Gaussian noise parameters, wherein the Gaussian noise parameters comprise at least one of a mean and a standard deviation;
acquiring a third degraded image by performing zoom blurring processing on the second degraded image according to a configured zoom blurring processing algorithm; and
acquiring a fourth degraded image by performing compression noise processing on the third degraded image according to a coding quality coefficient, wherein the coding quality coefficient is determined by:
Q
jepg
=
normal
ave
q
,
var
q
,
wherein Qjepg represents the coding quality coefficient, aveq represents a mean of values within a preset coefficient range, varq represents a variance of the values within the preset coefficient range, normal() represents a normal distribution function, and normal(aveq, varq) represents returning to one of a set of numbers that obeys the normal distribution of the mean aveq and the variance varq;
wherein each of the down-sampling processing algorithm and the zoom blurring processing algorithm comprises the following three sampling steps: a bicubic interpolation algorithm-based sampling step, a bilinear interpolation algorithm-based sampling step, and a Lanczos algorithm-based sampling step, wherein a sampling magnification of each of the sampling steps in the down-sampling processing algorithm is within a first magnification range, a sampling magnification of each of the sampling steps in the zoom blurring processing path is within a second magnification range, an upper limit of the first magnification range is less than or equal to 1, and an upper limit of the second magnification range is greater than 1.

### Claim 15 (depends on 1)

A method for training an image processing model, comprising:
training the image processing model by using a plurality of first original images and a plurality of degraded images corresponding to the plurality of first original images as a training data set, to enable a loss function of the image processing model to converge;
wherein the degraded images are acquired by processing the first original images according to the method for processing the image as defined in claim 1.

### Claim 16 (depends on 15)

A method for processing an image, comprising:
acquiring a second original image to be processed; and
acquiring a processed image output by an image processing model by inputting the second original image into the image processing model;
wherein the image processing model is acquired by training according to the method for training the image processing model as defined in claim 15.

### Claim 17 (depends on 1)

A device for processing an image, comprising:
a memory; and
a processor electrically connected to the memory, wherein
the memory stores a computer program executable by the processor to perform the method for processing the image as defined in claims 1.

### Claim 18 (depends on 15)

A device for processing an image, comprising:
a memory; and
a processor electrically connected to the memory, wherein
the memory stores a computer program executable by the processor to perform the method for training the image processing model as defined in claim 15.

### Claim 19 (depends on 16)

A device for processing an image, comprising:
a memory; and
a processor electrically connected to the memory, wherein
the memory stores a computer program executable by the processor to perform the method for processing the image as defined in claim 16.

### Claim 20 (depends on 1)

A non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, performs at least one of the method for processing the image as defined in claim 1 .
