Image analysis – Pattern recognition – Template matching
Reexamination Certificate
1999-04-13
2001-10-09
Bella, Matthew C. (Department: 2721)
Image analysis
Pattern recognition
Template matching
C382S209000, C382S278000
Reexamination Certificate
active
06301388
ABSTRACT:
This application is based on applications Nos. 10-102948 and 11-71022 filed in Japan, the contents of which are hereby incorporated by reference.
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to an image recognition technique, and in particular relates to an image matching technique for judging whether two plane images match.
2. Description of the Related Art
With rapid developments in hardware and software technologies in recent years, high-volume image data has become widely used in information processing.
For instance, an image database made up of considerable amounts of image data and managed under a database management system is used in various fields, such as medical care, office works, and scientific researches.
For instance, an image database made up of considerable amounts of image data and managed under a database management system is used in various fields, such as medical care, office work, and scientific research.
The last mentioned comparison method can also be applied to recognition of a captured image, by using the captured image as the retrieval key and searching the image database for an image that matches the captured image.
The following is an explanation of a conventional image matching technique used for the above comparison method.
FIGS. 1A and 1B
are a conceptual diagram showing image matching in the conventional technique.
Images
901
,
902
, and
903
are each made up of 32×32 sets of pixel data, while a part of pixel data which compose image
901
is enlarged in image
910
. Here, the sets of pixel data that make up each of the images have an 8-step monochrome graduation and therefore the actual images should be delivered in shades of gray, though the gray levels of the images are omitted in
FIG. 1. A
value assigned to each set of pixel data represents an intensity level of the pixel.
When images
902
and
903
are stored in an image database and image
901
is used as a retrieval key, image matching is performed as follows.
To compare image
901
with image
902
using the conventional image matching technique, the following calculation is performed for sets of pixel data C
1
, C
2
, C
3
, . . . , C
1024
which make up image
901
and sets of pixel data D
1
, D
2
, D
3
, . . . , D
1024
which make up image
902
.
&eegr;=(&Sgr;
C
j
·D
j
)/(&Sgr;
C
j
2
·&Sgr;D
j
2
) [
j=
1~1024]
When &eegr; approximates to 1, it is judged that images
901
and
902
match.
The same calculation is performed to compare image
901
with image
903
.
However, when an image has been captured using scaling and rotation, such a image is not congruent with but is merely similar to the other image with which the image is to be compared.
FIG. 2
shows two images
911
and
902
that differ in sizes and orientations.
To compare image
911
with image
902
using the conventional image matching technique, scaling and rotation have to be repeatedly performed on image
911
to judge whether image
911
matches image
902
, that is, whether the object in image
911
is identical to the object in image
902
.
SUMMARY OF THE INVENTION
The present invention aims to provide an image matching apparatus for comparing two images using a new image matching method, in view of diversifying needs for image matching with increasing uses of image data.
The above object can be fulfilled by an image matching apparatus for judging whether a first two-dimensional image and a second two-dimensional image match, the first two-dimensional image and the second two-dimensional image each being made up of a plurality of sets of pixel data in a two-dimensional xy coordinate plane, the image matching apparatus including: a first two-dimensional vector group storing unit for storing a plurality of two-dimensional vectors which are generated using a plurality of sets of pixel data of the first two-dimensional image and which respectively correspond to a plurality of positions in the first two-dimensional image, each of the plurality of two-dimensional vectors including a component relating to an x direction and a component relating to a y direction to show a feature, among a plurality of features of the first two-dimensional image, that is present in a corresponding position; a second two-dimensional vector group storing unit for storing a plurality of two-dimensional vectors which are generated using a plurality of sets of pixel data of the second two-dimensional image and which respectively correspond to a plurality of positions in the second two-dimensional image, each of the plurality of two-dimensional vectors including a component relating to the x direction and a component relating to the y direction to show a feature, among a plurality of features of the second two-dimensional image, that is present in a corresponding position; and a matching unit for judging that the first two-dimensional image and the second two-dimensional image match, when a distribution of the plurality of two-dimensional vectors stored in the first two-dimensional vector storing unit in the two-dimensional xy coordinate plane and a distribution of the plurality of two-dimensional vectors stored in the second two-dimensional vector storing unit in the two-dimensional xy coordinate plane satisfy predetermined matching conditions.
With this construction, a two-dimensional image can be expressed as image feature information about each of a plurality of positions in a two-dimensional xy coordinate plane, wherein the image feature information represents the change of values of a plurality of pixels around the position in the two-dimensional image. The image feature information shows, for example, spatial gradients of pixel data in each of the plurality of positions in the horizontal and vertical directions, where such spatial gradients are obtained by performing two-dimensional orthogonal wavelet decomposition on the two-dimensional image. The image feature information can be expressed as a two-dimensional vector composed of an x component and a y component.
By comparing image feature information of one two-dimensional image with image feature information of the other two-dimensional image, the image matching apparatus can perform image matching more efficiency than conventional techniques.
Also, when one of the plurality of positions is set as a reference position, a direction of a two-dimensional vector in the reference position is set as a reference direction. As a result, it is possible to describe each of the plurality of positions aside from the reference position by: a distance r from the reference position to the position; an angle &agr; which a two-dimensional vector from the reference position to the position forms with respect to the reference direction; and an angle &bgr; which a two-dimensional vector (image feature information composed of an x component and a y component) in the position forms with respect to the reference position. As a result, the two-dimensional image can be expressed using a group of vectors which are each made up of components derived from a distance r, an angle &agr;, and an angle &bgr;, without depending on whether the image has been rotated or not.
When two two-dimensional images that differ in orientations are each expressed as a vector group, the two images are judged as matching when angle components &agr; and &bgr; in a distribution of vectors of one vector group match angle components &agr; and &bgr; in a distribution of vectors of the other vector group to some extent. Likewise, when two two-dimensional images that differ in sizes and orientations are each expressed as a vector group, the two images are judged as matching when, in addition to a match of angle components &agr; and &bgr;, distance components r in a distribution of vectors of one vector group are proportional to components r in a distribution of vectors of the other vector group to some extent.
Thus, two two-dimensional images that differ in sizes and orientations can be compared with each other efficiently.
Here, a distribution of vectors denotes either
Bella Matthew C.
Desire Gregory
Matsushita Information Systems Co., Ltd.
Wenderoth , Lind & Ponack, L.L.P.
LandOfFree
Image matching 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 Image matching apparatus, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Image matching apparatus will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2554460