Image analysis – Pattern recognition
Reexamination Certificate
1998-07-08
2001-09-11
Johns, Andrew W. (Department: 2621)
Image analysis
Pattern recognition
Reexamination Certificate
active
06289123
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a character extracting method conducted in preparation for character recognition, and a character extracting apparatus for conducting the same. More particularly, the present invention relates to a character extracting method, which is used in, for example, a pre-process of character recognition for managing the quality of industrial products in the field of factory automation in an automated manner on a production line, and a character extracting apparatus for conducting the same. For example, the above-mentioned pre-process is a process for cutting out a character(s) on a wafer on a character-by-character basis by a character recognition apparatus for the purpose of quality management of the wafers produced.
2. Description of the Related Art
According to the character recognition technique in the field of office automation, a clear image can be obtained from a document of interest. Therefore, the character recognition technique in the office automation has been reached a substantially high-level recognition ratio.
According to the character recognition technique in the field of factory automation, however, a back-ground-light level varies depending upon the operation environment. Therefore, the background of an object of interest is less distinct, making the difference between a character(s) and the background unclear. Accordingly, the character recognition would be difficult without conducting a number of pre-processes prior to a character recognition process.
A general character recognition apparatus for the factory automation conducts a character recognition process by correctly cutting out a character region, i.e., a portion where a character is drawn, using the image processing technology. Provided that the character recognition process for the factory automation is the same as that for the office automation, whether the character recognition succeeds or not is determined by the character extracting process which is conducted as a pre-process of the character recognition process. When the character region can be correctly extracted, a character(s) can be recognized with a high recognition ratio, using a character recognition method similar to that used in the office automation.
A “method using region information” and a “method using edge information” are becoming popular as methods for extracting (cutting) a character region from an image.
One example of the “method using region information” is to set a threshold reflecting a local characteristic of an image I(i, j) (e.g., “Image Data Processing for Scientific Measurement” by Kawata et al., 1994, published by CQ Inc.).
The key to this threshold method is to properly select a threshold. A threshold T is given by, for example, the following expression (1):
T
=
1
N
∑
i
,
j
∑
CP
I
(
i
,
j
)
(
1
)
where P: a region;
N: the number of pixels in the region P;
I(i, j): a two-dimensional function representing an image; and
(i, j): a pixel position in coordinates.
According to the above expression (1), a region P centered around an pixel (i, j) is provided on a pixel-by-pixel basis, and a mean concentration value of each region P is set as a threshold T.
A Marr's zero-cross method (“Vision” by D. Marr, 1982, published by W. H. Freeman Inc.) is well known as a method using edge information. According to the zero-cross method, a Laplacian operation is applied to an original image according to the following expression (2), and the point where the operation result changes from positive to negative (i.e., zero-cross) is extracted as an edge of a character:
∇
2
f
≡
∂
2
f
∂
x
2
+
∂
2
f
∂
y
2
(
2
)
where ∇
2
: a Laplacian operator;
f: a two-dimensional function of x and y; and
(x, y): a pixel position in coordinates.
Alternatively, a method for first reducing the sharpness of the original image using a Gaussian function and then applying a Laplacian operation to the resultant image according to the following expression (3) is often used:
G(x, y)=1/(2&pgr;&sgr;
2
)exp(−(x
2
+y
2
)/(2&sgr;
2
));&sgr;>0(3)
where G(x, y): a two-dimensional Gaussian function; and
&sgr;: a spatial constant of the Gaussian function.
However, in the above-mentioned threshold method using region information, an importance level of the pixels in the region P is not considered on a pixel-by-pixel basis. In short, every pixel in the region P is regarded as being of the same importance. Moreover, every region P in the entire screen has the same size. Therefore, such a threshold T is not preferable. For such reasons as described above, a character(s) can not be precisely cut out by this threshold method. Consequently, such a high recognition ratio as obtained by the character recognition apparatus for office automation can not be expected.
SUMMARY OF THE INVENTION
According to one aspect of the present invention, a character extracting method includes the steps of: obtaining an image of a specific region; applying a Laplacian operation to the obtained image; binarizing the image resulting from the Laplacian operation; and separating and cutting a character on a character-by-character basis based on the binarized image.
In one example, in the binarizing step, in a case where a portion where a character to be cut out is drawn has a convex property, I
b
[i, j] is a first value when ∇
2
I
0
[i, j]≧0, and I
b
[i, j] is a second value when ∇
2
I
0
[i, j]<0; and in a case where the portion where the character to be cut out has a concave property, I
b
[i, j] is the first value when ∇
2
I
0
[i, j]≦0, and I
b
[i, j] is the second value when ∇
2
I
0
[i, j]>0, where ∇
2
represents a Laplacian operator; [i, j] represents a pixel position in the image; I
0
[i, j] represents a pixel value corresponding to the pixel position [i, j]; and I
b
[i, j] represents a binarized pixel-value corresponding to a pixel value I
0
[i, j].
In one example, the first value corresponds to a level
0
, and the second value corresponds to a level
255
.
In one example, the convex property indicates such a property that the portion where the character to be cut out is drawn becomes brighter toward the center thereof, and the concave property indicates such a property that the portion where the character to be cut out is drawn becomes darker toward the center thereof.
In one example, a character extracting method further includes, after the binarizing step, the step of: filtering noise from the binarized image.
In one example, the step of filtering the noise from the binarized image uses an arbitrary pixel and a pixel adjacent to the arbitrary pixel in the binarized image to filter the noise from the arbitrary pixel.
In one example, the step of filtering the noise from the binarized image filters the noise by applying a minimum filter to a region having the arbitrary pixel and the adjacent pixel.
In one example, the step of filtering the noise from the binarized image filters the noise by applying an AND operation to a region having the arbitrary pixel and the adjacent pixel.
In one example, the step of filtering the noise from the binarized image filters the noise by applying a mean-value operation to a region having the arbitrary pixel and the adjacent pixel.
In one example, a character extracting method further includes, after the step of filtering the noise from the binarized image, the step of conducting an expansion process and/or a contraction process in order to shape a deformed character and/or to fill a hole which is present in the portion where the character to be cut out is drawn.
In one example, a character extracting method further includes, prior to the step of conducting the expansion process and/or the contraction process, the step of determining an outer frame of the character to be cut out.
In one example, a
Hatakeyama Koichi
Xiaomang Zhang
Azarian Seyed
Johns Andrew W.
Sharp Kabushiki Kaisha
LandOfFree
Character extracting method and its apparatus does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Character extracting method and its apparatus, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Character extracting method and its apparatus will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2523785