# US 2012/0250105 A1 · Method Of Analyzing Digital Document Images

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

- Number: US 2012/0250105 A1
- Document ID: US20120250105A1
- Status: Published application
- Kind: A1
- Application number: 13/075,978
- Filed: 2011-03-30
- Published: 2012-10-04
- Earliest claimed: 2011-03-30
- Assignees: Seiko Epson Corporation
- Inventors: Rastislav Lukac
- Web page: https://patentgrep.com/patent/US20120250105A1

## CPC

- G06V 30/414: Extracting the geometrical structure, e.g. layout tree; Block segmentation, e.g. bounding boxes for graphics or text
- G06V 30/162: Quantising the image signal
- G06V 30/10: Character recognition
- G06V 30/147: {Determination of region of interest}
- G06V 30/1607: {Correcting image deformation, e.g. trapezoidal deformation caused by perspective}
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

Analyzing an input image, the input image being one of a digitized image stored in a memory or a scanned image from a scanner. Forming a feature image from the input image by dividing the input image into a plurality of blocks of pixels, thus associating each block of pixels in the input image with a single pixel in the feature image, and outputting the feature image for further analysis or storage in a memory. Example embodiments extract and analyze features from a document image to detect particular characteristics associated with the page area, the distortion area, and the book spine area. Extracted features can be further analyzed to detect document characteristics at the paragraph, line, word, and character levels.

## Claims

### Claim 1

A method for analyzing an input image x with K1×K2 pixels x(r,s) where (r,s) denotes the pixel location with r=1, 2, . . . , K1 indicating the image row, and s=1, 2, . . . , K2 indicating the image column, the input image being one of a digitized image stored in a memory or a scanned image from a scanner, comprising:
using a processor to form a feature image z from the input image x by:
dividing the input image x into a plurality of blocks of pixels each block having a block size of g1×g2 pixels, where g1 denotes a number of image rows in a block and g2 denotes a number of image columns in a block,
associating each block of pixels in the input image x with a single pixel in the feature image z, with the feature image z consisting of K1/g1×K2/g2 pixels, and
outputting the feature image z for further analysis or storage in a memory,
wherein the feature image z is a two-channel image with feature pixels z(m,n)=[f(m,n),b(m,n)], for m=1, 2, . . . , K1/g1 and n=1, 2, . . . , K2/g2, wherein f(m,n) and b(m,n) denote, respectively, the foreground and background components of a feature pixel z(m,n),
wherein the foreground and background components of a feature pixel z(m,n) are respectively defined as follows:
f(m,n)=min{x(r,s);(m−1)g1<r≦mg1,(n−1)g2<s≦ng2}
b(m,n)=max{x(r,s);(m−1)g1<r≦mg1,(n−1)g2<s≦ng2}
where min and max are the minimum and maximum operators, and
wherein the processor detects pages according to:
d
(
m
,
n
)
=
{
1
for


b
(
m
,
n
)
-
f
(
m
,
n
)
<
θ


and


b
(
m
,
n
)
>
ϑ
0
otherwise
.
where θ and φ are tunable parameters used to evaluate a similarity between a foreground and a background in input image x and identify pixels with high contributions of background and foreground components, and wherein d(m,n) is a pixel in a binary map d of dimensions identical to that of the feature image z, and the value d(m,n)=1 indicates that the location (m,n) in the feature image z corresponds to a page area.

### Claim at position 2

2-4. (canceled)

### Claim 5 (depends on 1)

A method as in claim 1, wherein the processor detects pages according to:
d
(
m
,
n
)
=
{
1
for


f
(
m
,
n
)
>
β
0
otherwise
where β is the average value of b(m,n) components which satisfy b(m,n)−f(m,n)<φ and b(m,n)>φ, and, wherein d(m,n) is a pixel in a binary map d of dimensions identical to that of the feature image z, and wherein the value d(m,n)=1 indicates that the location (m,n) in the feature image corresponds to a page area.

### Claim 6 (depends on 1)

A method as in claim 1, wherein the processor subjects the binary map d to object segmentation by grouping adjacent pixels with d(m,n)=1.

### Claim 7 (depends on 6)

A method as in claim 6, wherein the processor partitions the binary map d into N disjoint objects Oi={(m,n)εΦi;d(m,n)i=1}, for i=1, 2, . . . , N, wherein each object is characterized by Φi which is the set of pixel locations (m,n) where d(m,n)i=1, and a Φiy×Φix is a bounding box with height Φiy and width Φix.

### Claim 8 (depends on 7)

A method as in claim 7, wherein the processor removes small objects and objects with bounding boxes of irregular aspect ratios as follows:
O
i
=
{
{
d
(
m
,
n
)
i
=
0
;
(
m
,
n
)
∈
Φ
i
}
if


Φ
i
y
<
T
y


or


Φ
i
x
<
T
x


or
max

(
Φ
i
y
,
Φ
i
x
)
/
min

(
Φ
i
y
,
Φ
i
x
)
>
T
O
i
otherwise
where Ty and Tx denote a minimum height and width, respectively of an object and T is a maximum aspect ratio of the object.

### Claim 9 (depends on 7)

A method as in claim 7 wherein the processor analyzes the objects to detect a page orientation in the input image x by comparing a height of the object to a width of the object.

### Claim 10 (depends on 7)

A method as in claim 7 wherein the processor analyzes the objects to detect a book spine in the input image x.

### Claim 11

A device for analyzing an input image x with K1×K2 pixels x(r,s) where (r,s) denotes the pixel location with r=1, 2, . . . , K1 indicating the image row, and s=1, 2, . . . , K2 indicating the image column, comprising:
an image capture unit that captures input image x;
a memory that stores input image x; and
a processor that forms a feature image z from the input image x by:
dividing the input image x into a plurality of blocks of pixels each block having a block size of g1×g2 pixels, where g1 denotes a number of image rows in a block and g2 denotes a number of image columns in a block,
associating each block of pixels in the input image x with a single pixel in the feature image z, with the feature image z consisting of K1/g1×K2/g2 pixels, and
outputting the feature image z for further analysis or storage in a memory,
wherein the feature image z is a two-channel image with feature pixels z(m,n)=[f(m,n),b(m,n)], for m=1, 2, . . . , K1/g1 and n=1, 2, . . . , K2/g2, wherein f(m,n) and b(m,n) denote, respectively, the foreground and background components of a feature pixel z(m,n),
wherein the foreground and background components of a feature pixel z(m,n) are respectively defined as follows:
f(m,n)=min{x(r,s);(m−1)g1<r≦mg1,(n−1)g2<s≦ng2}
b(m,n)=max{x(r,s);(m−1)g1<r≦mg1,(n−1)g2<s≦ng2}
where min and max are the minimum and maximum operators, and
wherein the processor detects pages according to:
d
(
m
,
n
)
=
{
1
for


b
(
m
,
n
)
-
f
(
m
,
n
)
<
θ


and


b
(
m
,
n
)
>
ϑ
0
otherwise
where θ and φ are tunable parameters used to evaluate a similarity between a foreground and a background in input image x and identify pixels with high contributions of background and foreground components, and wherein d(m,n) is a pixel in a binary map d of dimensions identical to that of the feature image z, and the value d(m,n)=1 indicates that the location (m,n) in the feature image z corresponds to a page area.

### Claim at position 10

12-14. (canceled)

### Claim 15 (depends on 11)

A device as in claim 11, wherein the image capture unit is a scanning unit.

### Claim 16

One or more tangible computer-readable media having computer-readable instructions thereon, which, when executed by a processor analyze an input image x with K1×K2 pixels x(r,s) where (r,s) denotes the pixel location with r=1, 2, . . . , K1 indicating the image row, and s=1, 2, . . . , K2 indicating the image column, the input image being one of a digitized image stored in a memory or a scanned image from a scanner, wherein:
the processor forms a feature image z from the input image x by:
dividing the input image x into a plurality of blocks of pixels each block having a block size of g1×g2 pixels, where g1 denotes a number of image rows in a block and g2 denotes a number of image columns in a block,
associating each block of pixels in the input image x with a single pixel in the feature image z, with the feature image z consisting of K1/g1×K2/g2 pixels, and
outputting the feature image z for further analysis or storage in a memory,
wherein the feature image z is a two-channel image with feature pixels z(m,n)=[f(m,n),b(m,n)], for m=1, 2, . . . , K1/g1 and n=1, 2, . . . , K2/g2, wherein f(m,n) and b(m,n) denote, respectively, the foreground and background components of a feature pixel z(m,n),
wherein the foreground and background components of a feature pixel z(m,n) are respectively defined as follows:
f(m,n)=min{x(r,s);(m−1)g1<r≦mg1,(n−1)g2<s≦ng2}
b(m,n)=max{x(r,s);(m−1)g1<r≦mg1,(n−1)g2<s≦ng2}
where min and max are the minimum and maximum operators, and
wherein the processor detects pages according to:
d
(
m
,
n
)
=
{
1
for


b
(
m
,
n
)
-
f
(
m
,
n
)
<
θ


and


b
(
m
,
n
)
>
ϑ
0
otherwise
where θ and φ are tunable parameters used to evaluate a similarity between a foreground and a background in input image x and identify pixels with high contributions of background and foreground components, and wherein d(m,n) is a pixel in a binary map d of dimensions identical to that of the feature image z, and the value d(m,n)=1 indicates that the location (m,n) in the feature image z corresponds to a page area.

### Claim at position 13

17-19. (canceled)

### Claim 20 (depends on 16)

The one or more tangible computer-readable media as in claim 16, wherein the processor detects pages according to:
d
(
m
,
n
)
=
{
1
for


f
(
m
,
n
)
>
β
0
otherwise
where β is the average value of b(m,n) components which satisfy b(m,n)−f(m,n)<θ and b(m,n)>φ, and, wherein d(m,n) is a pixel in a binary map d of dimensions identical to that of the feature image z, and wherein the value d(m,n)=1 indicates that the location (m,n) in the feature image corresponds to a page area.
