Image analysis – Color image processing – Image segmentation using color
Reexamination Certificate
2000-08-21
2004-06-29
Mehta, Bhavesh M. (Department: 2625)
Image analysis
Color image processing
Image segmentation using color
C382S173000, C382S266000
Reexamination Certificate
active
06757427
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to an image region determining apparatus with an image region determining function.
2. Description of the Related Art
A digital image is frequently process through the spread of the image processing apparatus in these days. As a method of compressing such digital image data, a DCT coding method such as the JPEG method is widely used generally.
In accordance with the above-mentioned DCT coding method, an image is resolved based on the frequency component strength and a high frequency component strength is quantized roughly, and data compression is carried out. Therefore, the DCT coding system of the JPEG system is an optimal compressing method to a natural image, in which the picture element values change continuously.
However, in an actual case, an image contains characters and illustrations or drawings in addition to the natural image manipulated as a photograph object (reading object). The characters and drawings are superposed on the natural image, which is seen on the cover of a magazine. When the DCT coding method is carried out to such a natural image to compress and code the image, there is a problem that moskeyte noize is generated around the contour portion of regions of the characters and the drawing. As a result, the degradation of image quality is caused to a large extent. The reason is in that the data of the high frequency component strength of the image has been reduced so that the steep edge can not be reproduced in the contour portion of the regions of the characters and the drawings.
As the technique for solving such a problem, a method is known in which the quantization step of the DCT coefficient is controlled. In this technique, the block which contains the contour portion of the character region or drawing region can be coded in a high precision, so that a coding warp in the contour portion is improved.
However, in the method in which the above quantizing step is controlled, when the character region and the drawing region occupy a large portion of the image, there is a problem that a code quantity increases. This is because the most of the blocks are coded in the high precision to attain an enough image quality.
On the other hand, it is known that an image is divided into two regions of the character region and the natural image region, and each region is compressed by the different coding method. In accordance with this method, an entropy coding method such as a run length method is used for the character region, and a conversion coding method such that DOT and a wavelet are used for the natural image region is used.
In the above method, because each divided region (the character region and the natural image region) can be compressed in the method of coding suitable for an image characteristic for every region, the coding quality can be improved, compared with the case that the JPEG method is independently used. Also, the increase of the code quantity can be held even when the rate of the character region is increased. However, there is a problem of limitations in the processing quantity of the region separation, the size and the shape of the character which can be separated.
As the conventional example for solving the above-mentioned problem, a conventional image processing apparatus with a region determining function is shown in FIG.
1
. In the conventional determining process of the character region and the photograph region in the color manuscript, the image data of the color manuscript is converted into RGB color space data represented by 3 primary colors of R (red), G (green) and B (the blue). A determination of the character region and the photograph region in the input image is carried out using the converted RGB data.
The manuscript is generally read by a CCD device or a scanner in the form of RGB data, and then each of the 3 primary color data (R, G, B) is A/D-converted. A character region and a photograph region in the input image are determined using the 8-bit data. Such a technique is known. Hereinafter, the technique will be described with reference to FIG.
1
.
Referring to
FIG. 1
, the conventional image region determining apparatus is mainly composed of a color reducing and quantizing section
101
, a constant color region clustering section
102
, a color deviation variance detecting section
103
, an edge-in-window detecting section
104
, a region determining section
105
, an image processing section
106
and a data compressing section
107
.
In the above conventional image region determining apparatus, each of the RGB data of the input image is expressed in 8 bits. Such data are inputted to the color reducing and quantizing section
101
, and a color reducing process is carried out therein to reduce 1 to 3 bits of the data. As this color reducing process, there is a method of simply cutting off lower 5 to 7 bits of the RGB data to be inputted and there is a method of rounding. For example, when the lower 5 bits of each 8-bit data of the RGB data are simply cut off, each of the RGB data becomes 3-bit data. At this time, the number of reduced colors is 512 (=8×8×8). The region clustering process is carried out to these 512 colors, and the region determination is sufficient to be carried out for every clustered region.
In the constant color region clustering section
102
, an optional concerned picture element is compared with eight picture elements which are neighbor to the concerned picture element, as shown in
FIG. 2
using the color reduced data obtained through the cutting off or rounding of the lower bits by the color reducing and quantizing section
101
. When it is determined to be the same quantization value based on the comparing result, an integrating process is carried out as a constant color region. For example, as shown in
FIG. 2
, the concerned picture element is a concerned picture element P. It is supposed that eight picture elements as the peripheral picture elements which surround the concerned picture element P are a picture element P
1
to a picture element P
8
. The picture element value of the concerned picture element P is compared with the picture element value of each of picture elements P
1
to P
8
in order to integrate the picture elements with respect to the same quantization value. Thus, the clustering process is carried out as the same color region.
In this case, in the constant color region clustering section
102
, a region after the clustering process is composed of one constant color region or several constant color regions in the character regions and the drawing regions in which color variation is regarded to be less in the color manuscript. The characteristic quantity is found for every region to which the clustering process is carried out so that a region determination is carried out. In the conventional example, the color deviation is detected for every region in the region to which the clustering process is carried out in the color deviation variance detecting section
103
. The edge portion or boundary portion of the character region in the window is detected by the edge-in-window detecting section
104
. The determination of the character region or the photograph region is carried out in the region determining section
105
using the detection data.
In the detecting process in the color deviation variance detecting section
103
, it is supposed that the region composed of a plurality of picture elements and integrated to have the same quantization value by the constant color region clustering section
102
is a constant color region A. In this case, the color deviation in the constant color region A is calculated using the following equation (1) shown below:
V
(
A
)=(1/
N
(
A
))×&Sgr;(Dif(
C
(
P
),
C
(Aav.))):
n
(
P&egr;A
) (1)
where in the above-mentioned equation (1), V(A) is the variance in the color region A (deviation degree of color), N(A) is the number of picture elements in the color region A, C(P) is each 8-bit data value of RGB data in the optional co
Hung Yubin
Mehta Bhavesh M.
LandOfFree
Edge enhancement preprocessing with image region determining... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Edge enhancement preprocessing with image region determining..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Edge enhancement preprocessing with image region determining... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3302904