Image analysis – Color image processing – Pattern recognition or classification using color
Reexamination Certificate
1999-01-12
2004-03-02
Boudreau, Leo (Department: 2621)
Image analysis
Color image processing
Pattern recognition or classification using color
C382S225000
Reexamination Certificate
active
06701010
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a color image processing apparatus and a pattern extracting apparatus, in particular, to those suitable for extracting a title or the like from a color image.
2. Description of the Related Art
In recent years, as computers and their peripheral units such as printers have become common and their costs have been decreased, color images have been used in a variety of fields. Thus, a technology for dividing a color image into several areas and extracting only a particular area has been desired. For example, a technology for extracting areas with the same color from a color image has been desired. When a color scenery image photographed by a CCD camera is used for an input image to be process, such a technology has been widely desired in many applications for selecting fruit and watching cars and people for securities.
When a color document image is used as an input image, such a technology is expected to automatically extract a document name and a keyword from the image. Examples of such a technology are data retrieval systems such as book categorizing systems in libraries and automatic management systems. In addition, such a technology is used for automatically assigning keywords and file names in groupware for storing and sharing image data as a database. Such information can be used for retrieving desired data from a large number of color document images.
As technologies for extracting a particular area from a color image, the following methods have been proposed.
(1) RGB Color Space Clustering Method
In the RGB color space clustering method, a color separated image is generated. In other words, pixels of an RGB image photographed by a CCD camera are clustered in the RGB space. With pixels in one cluster, an image of each color is generated. Thus, areas with the same color are extracted. Areas generated in such a method are combined so as to extract a new area.
FIG. 1
is a schematic diagram for explaining the conventional RGB color space clustering method.
In
FIG. 1
, assuming that a color document image
501
is input, patterns with similar colors are extracted and clustered. For example, assuming that patterns
502
and
507
are blue group colors, a pattern
503
is a green group color, and patterns
504
to
506
are red group colors, a cluster
508
that includes the blue group color patterns
502
and
507
, a cluster
509
that includes the green group color pattern
503
, and a cluster
510
that includes the red group color patterns
504
to
506
are generated in the RGB space.
When the clusters
508
to
510
are generated, images of the clusters
508
to
510
are generated with pixels that are included therein. Thus, for the cluster
508
, a color separated image
501
a
composed of patterns
502
′ and
507
′ is generated. For the cluster
509
, a color separated image
501
b
composed of a pattern
503
′ is generated. For the cluster
510
, a color separated image
501
c
composed of patterns
504
′ to
506
′ is generated.
(2) Non-RGB Color Space Clustering Method
All pixels of an RGB image represented in the RGB space are converted into another color space such as HSV. The pixels are clustered in the color space in a particular manner. Images are generated with pixels that are included in each cluster so as to extract areas with the same colors. The obtained areas are combined so as to extract a new area.
The following technical papers that describe technologies for extracting character areas from color document images are known.
Senda et. al., “Method for extracting a character pattern from a color image due to a single color of characters (translated title)”, The Institute of Electronics, Information and Communication Engineers, Japan, PRU 94-04, pp 17-24,
Uehane et. al., “Extracting a character area from a color image using iso-color line process (translated title)”, The Institute of Electronics, Information and Communication Engineers, Japan, PRU 94-09, pp 9-16,
Matsuo et. al., “Extracting a single color character area from a color document image (translated title)”, 1997 Annual Convention, The Institute of Electronics, Information and Communication Engineers, Japan, D-12-19,
Matsuo et. al., “Extracting a character string from a scenery image with gradation and color information (translated title)”, The Institute of Electronics, Information and Communication Engineers, Japan, PRU 92-121, pp 25-32.
(3) Area Expanding Method
In the area expanding method, adjacent pixels are assigned labels corresponding to only similarities of colors. In other words, the maximum value (max) of each color element of (R, G, B) of pixels represented with RGB is obtained and normalized as (R/max, G/max, B/max). Thus, a normalized image is generated.
FIG. 2
is a schematic diagram showing the conventional area expanding method.
In
FIG. 2
, pixels P
1
, P
2
, and so forth in an image are represented with color elements RGB as PL(R
1
, G
1
, B
1
), P
2
(R
2
, G
2
, B
2
), and so forth [1].
Next, the maximum value of each color element is obtained. For example, the maximum value of R
1
, G
1
, and B
1
of a pixel P
1
is denoted by maxi. Likewise, the maximum value of R
2
, G
2
, and B
2
of a pixel P
2
is denoted by max
2
. With the maximum values, each color element is normalized. Thus, normalized pixels P
1
′ (R
1
/max
1
, G
1
/max
1
, B
1
/max
1
) and P
2
′(R
2
/max
2
, G
2
/max
2
, B
2
/max
2
) are obtained [2].
The square of the difference of each color element of the normalized pixels P
1
′ and P
2
′ is obtained. The results are cumulated so as to obtain the distance between the adjacent pixels P
1
′ and P
2
′ as d=(R
1
/max
1
−R
2
/max
2
)
2
+(G
1
/max
1
−G
2
/max
2
)
2
+(B
1
/max
1
−B
2
/max
2
)
2
[3].
Thus, when the distance d is smaller than a predetermined fixed threshold value THO, the pixels P
1
and P
2
are treated as those with the same color and assigned the same label. After all the image is assigned labels, the same color areas with the same labels are extracted.
In the area expanding method, since only adjacent pixels are processed, the process time of this method is shorter than that of the RGB color space clustering method.
For details of the area expanding method, refer to Japanese Patent Laid-Open Publication No. 5-298443.
In addition, as a method for extracting a character area from a color separated image, the above-mentioned method (by Uehane et. al, “Extracting a character area from a color image using iso-color line process”, The Institute of Electronics, Information and Communication Engineers, Japan, PRU 94-09, pp 9-16) is known. In this method, a character area is extracted from a color image with the following steps.
Enclosing rectangles in connected areas are obtained from a single-color separated image.
Enclosing rectangles are limited in a predetermined range of the sizes and shapes thereof.
An adjacent rectangle search range of each rectangle is assigned. Rectangles are searched as a group in each search range.
Rectangles with a high linearity of center of gravity are kept in each group.
An enclosing rectangle of each group is obtained and a pattern with a color similar to a color of the area that composed the group is extracted.
However, the conventional clustering method for clustering the same color area of a color image has the following problems.
In the RGB color space clustering method or another color space clustering method (for example, HSV space clustering method), all pixels of the image are clustered. Thus, even if the color of the pattern
502
is different from the color of the pattern
507
, when their colors are similar to each other, the patterns
502
and
507
may be categorized as the same cluster
508
. In this case, due to the color of the pattern
507
, the shape of the cluster of the color of the pattern
502
is varied. Thus, the color range of the cluster of the color of the pattern
502
is distorted.
Boudreau Leo
Fujitsu Limited
Sherali Ishrat
Staas & Halsey , LLP
LandOfFree
Color image processing apparatus and pattern extracting... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Color image processing apparatus and pattern extracting..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Color image processing apparatus and pattern extracting... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3268907