# US 2023/0156167 A1 · Image recognition system and training method therefor

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

- Number: US 2023/0156167 A1
- Document ID: US20230156167A1
- Status: Published application
- Kind: A1
- Application number: 17/716,554
- Filed: 2022-04-08
- Published: 2023-05-18
- Earliest claimed: 2021-11-09
- Assignees: Realtek Semiconductor Corp.
- Applicants: Realtek Semiconductor Corp.
- Inventors: Wei-Hsiang Shen; Kai-Chun Su; Tsung-Hsuan Li; Chao-Hsun Yang; Shih-Tse Chen
- Web page: https://patentgrep.com/patent/US20230156167A1

## CPC

- H04N 9/43: Conversion of monochrome picture signals to colour picture signals for colour picture display
- H04N 9/64: Circuits for processing colour signals (H04N9/77 takes precedence; camera processing pipelines for processing colour signals H04N23/84)
- G06T 7/74: {involving reference images or patches}
- G06T 7/90: Determination of colour characteristics
- G06T 11/10: Texturing; Colouring; Generation of textures or colours (retouching, inpainting or scratch removal G06T5/77)
- G06V 10/56: relating to colour
- G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06V 10/7747: {Organisation of the process, e.g. bagging or boosting}
- G06V 10/82: using neural networks
- H04N 5/33: Transforming infrared radiation (cameras or camera modules for generating image signals from infrared radiation H04N23/20; circuitry of SSIS for transforming infrared radiation into image signals H04N25/20)
- G06T 2207/10024: Color image
- G06T 2207/10048: Infrared image
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06V 2201/07: Target detection
- H04N: PICTORIAL COMMUNICATION, e.g. TELEVISION
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

An image recognition system includes a color conversion module and a target recognition module. The color conversion module is configured to convert a gray-level image into a preset color image according to a conversion function. The target recognition module includes a machine learning algorithm, and the machine learning algorithm includes a plurality of functions and a plurality of parameters. The machine learning algorithm receives the preset color image, and outputs a recognition result according to the functions and the parameters, the recognition result including an existent target or a null target.

## Claims

### Claim 1

An image recognition system, comprising:
a color conversion module, configured to convert a gray-level image into a preset color image according to a conversion function; and
a target recognition module, comprising a machine learning algorithm, wherein the machine learning algorithm comprises a plurality of functions and a plurality of parameters, the machine learning algorithm receives the preset color image, and outputs a recognition result according to the functions and the parameters, the recognition result comprising an existent target or a null target.

### Claim 2 (depends on 1)

The image recognition system according to claim 1, wherein the gray-level image comprises brightness information, the conversion function is used for converting the brightness information into first color information, and the conversion function is:
I
R
=
f
W
R
×
I
b
r
i
g
h
t
n
e
s
s
+
b
R
I
G
=
f
W
G
×
I
b
r
i
g
h
t
n
e
s
s
+
b
G
I
B
=
f
W
B
×
I
b
r
i
g
h
t
n
e
s
s
+
b
B
I
R
G
B
=
I
R
+
I
G
+
I
B
wherein Ibrightness is the brightness information, WR is a red weight, WG is a green weight, WB is a blue weight, bR is a red offset, bG is a green offset, bB is a blue offset, IR is red color information, IG is green color information, IB is blue color information, IRGB is the first color information, and f() is a limited output function, an output range of the limited output function being between a value 0 and a value 255.

### Claim 3 (depends on 2)

The image recognition system according to claim 2, wherein the blue offset (bB), the green offset (bG), and the red offset (bR) are negative values, an absolute value of the blue offset (bB) is greater than an absolute value of the green offset (bG), and the absolute value of the green offset (bG) is greater than an absolute value of the red offset (bR).

### Claim 4 (depends on 3)

The image recognition system according to claim 3, wherein a difference between the green offset (bG) and the red offset (bR) is the same as a difference between the blue offset (bB) and the green offset (bG).

### Claim 5 (depends on 2)

The image recognition system according to claim 2, wherein the red weight WR, the green weight WG, and the blue weight WB are the same.

### Claim 6 (depends on 5)

The image recognition system according to claim 5, wherein the red weight WR, the green weight WG, and the blue weight WB are 3, the red offset (bR) is 0, the green offset (bG) is -1 × 255, and the blue offset (bB) is -2 × 255.

### Claim 7 (depends on 2)

The image recognition system according to claim 2, wherein the limited output function f() is:
f
x
=
Re
L
U
x
−
Re
L
U
x
−
255
wherein ReLU is a rectified linear unit, and x is an input value of the limited output function (f).

### Claim 8 (depends on 2)

The image recognition system according to claim 2, further comprising: converting the first color information into second color information according to the following formulas:
I
Y
I
U
I
V
=
0.299
0.587
0.114
−
0.169
−
0.331
0.5
0.5
−
0.419
−
0.081
I
R
I
G
I
B
+
0
128
128
I
Y
U
V
=
I
Y
+
I
U
+
I
V
wherein IY is the brightness information, IU is first chrominance information, IV is second chrominance information, and IYUV is the second color information.

### Claim 9 (depends on 1)

The image recognition system according to claim 1, wherein the gray-level image comprises brightness information, the conversion function is used for converting the brightness information into first color information, and the conversion function is:
I
R
=
255
×
I
b
r
i
g
h
t
n
e
s
s
255
γ
R
I
G
=
255
×
I
b
r
i
g
h
t
n
e
s
s
255
γ
G
I
B
=
255
×
I
b
r
i
g
h
t
n
e
s
s
255
γ
B
I
R
G
B
=
I
R
+
I
G
+
I
B
wherein Ibrightness is the brightness information, IR is red color information, IG is green color information, IB is blue color information, γR is a red gamma value, γG is a green gamma value, γB is a blue gamma value, and IRGB is the first color information.

### Claim 10 (depends on 9)

The image recognition system according to claim 9, wherein the blue gamma value (γB) is greater than the green gamma value (γG), and the green gamma value (γG) is greater than the red gamma value (γR).

### Claim 11 (depends on 10)

The image recognition system according to claim 10, wherein a difference between the green gamma value (γG) and the red gamma value (γR) is the same as a difference between the blue gamma value (γB) and the green gamma value (γG).

### Claim 12 (depends on 10)

The image recognition system according to claim 10, wherein the red gamma value (γR) is between 0 and 1, the green gamma value (γG) is 1, and the blue gamma value (γB) is greater than 1.

### Claim 13 (depends on 1)

The image recognition system according to claim 1, further comprising a color recognition module, the color recognition module being configured to:
receive image data;
determine whether the image data is the gray-level image according to the image data or a photographing parameter; and
transmit the gray-level image to the target recognition module.

### Claim 14 (depends on 13)

The image recognition system according to claim 13, wherein when the color recognition module determines that the image data is not the gray-level image, the color recognition module is further configured to transmit the image data to the target recognition module, and the machine learning algorithm is further used for receiving the image data, and outputting the recognition result according to the functions and the parameters.

### Claim 15 (depends on 1)

The image recognition system according to claim 1, wherein the image recognition system is a gray-level image recognition system, configured to control a photographing module to capture the gray-level image in an infrared light sensing mode, and transmit the gray-level image to the color conversion module.

### Claim 16

A training method for an image recognition system, comprising:
reading an original color image, wherein the original color image comprises a training object, and the training object corresponds to a training tag;
converting the original color image into a gray-level image;
converting the gray-level image into a preset color image according to a conversion function; and
adjusting parameters of a machine learning algorithm according to the preset color image and the training tag.

### Claim 17 (depends on 16)

The method according to claim 16, further comprising:
reading an original gray-level image, wherein the original gray-level image comprises another training object, and the another training object corresponds to another training tag;
converting the original gray-level image into another preset color image according to the conversion function; and
adjusting the parameters of the machine learning algorithm according to the another preset color image and the another training tag.

### Claim 18 (depends on 16)

The method according to claim 16, further comprising:
adjusting the parameters of the machine learning algorithm according to the original color image and the training tag.

### Claim 19 (depends on 16)

The method according to claim 16, wherein the gray-level image comprises brightness information, the conversion function is used for converting the brightness information into first color information, and the conversion function is:
I
R
=
f
W
R
×
I
b
r
i
g
h
t
n
e
s
s
+
b
R
I
G
=
f
W
G
×
I
b
r
i
g
h
t
n
e
s
s
+
b
G
I
B
=
f
W
B
×
I
b
r
i
g
h
t
n
e
s
s
+
b
B
I
R
G
B
=
I
R
+
I
G
+
I
B
wherein Ibrightness is the brightness information, WR is a red weight, WG is a green weight, WB is a blue weight, bR is a red offset, bG is a green offset, bB is a blue offset, IR is red color information, IG is green color information, IB is blue color information, IRGB is the first color information, and f() is a limited output function, an output range of the limited output function being between a value 0 and a value 255.

### Claim 20 (depends on 19)

The method according to claim 19, wherein the limited output function f() is:
f
x
=
Re
L
U
x
−
Re
L
U
x
−
255
wherein ReLU is a rectified linear unit, and x is an input value of the limited output function (f).
