Image analysis – Pattern recognition – Feature extraction
Reexamination Certificate
2000-03-30
2003-11-04
Mariam, Daniel G. (Department: 2621)
Image analysis
Pattern recognition
Feature extraction
C382S173000, C382S180000, C382S190000, C382S205000, C382S305000, C358S403000
Reexamination Certificate
active
06643400
ABSTRACT:
This application is based on applications Nos. 11-92050 and 11-92094 filed in Japan, the contents of which are incorporated herein by reference.
BACKGROUND OF THE INVENTION
1. Technical Field of the Invention
The present invention relates to an image processing technique for recognizing a specific pattern included in an image.
2. Related Art
Conventionally, in an image processing apparatus, the position of a specific pattern included in an image has been detected and a pattern matching processing has been carried out to recognize the shape of the specific pattern by storing data on each pixel fetched into a camera or the like as binary data or multivalued data in an image memory and matching, for each pixel, the stored data with data on each pixel of a reference pattern stored beforehand in the memory. In this matching, the whole image has been scanned while shifting the reference pattern every pixel for an image including the specific pattern in the x or y direction of the image.
In this method, however, the image including the specific pattern and the reference pattern are shifted and matched every pixel. Therefore, there has been a problem in that a very long time is required for the process.
In the image recognizing process, moreover, examples of a method for detecting the position of the specific pattern of an image implementing a high speed include a method for creating a reduced image from the image, performing a pattern matching on the reduced image to retrieve the specific pattern, and carrying out retrieval in the region of an original image corresponding to the detecting position of the specific pattern on the reduced image again. In this method, however, the specific pattern cannot be detected accurately depending on an algorithm for creating the reduced image in some cases. For example, in the case of the image having a pixel value which is alternately different for each pixel such as checkers, when thinning is carried out every other pixel to create a reduced image, the reduced image thus created has only one of pixel values appearing on an original image. Consequently, a correct specific pattern cannot be detected from such a reduced image.
Moreover, examples of a process to be carried out prior to a process such as image recognition include a labeling process of detecting a region to be formed by pixels having the same value which are continuous in a binary image (hereinafter referred to as a “connecting region”) and giving a label number to the connecting region. As a method for the labeling process, various methods have conventionally been proposed.
FIG. 21
is a diagram illustrating one example of the labeling processes in the connecting region of the image according to the prior art. In
FIG. 21
, an original image
20
includes three connecting regions A, B and C. These connecting regions A, B and C are continuous regions comprising pixels having pixel value of “1”. When labeling the connecting region of the image, the whole image is scanned to examine the pixel value of each pixel and to give a provisional label number for a pixel decided to constitute the connecting region. At this time, the provisional label number is not always accurate. Therefore, re-labeling of the label number is carried out later in some cases. For this reason, information for the re-labeling is stored in tables
51
and
53
. Then, an image is scanned again to replace the provisional label number with a real label number with reference to the tables
51
and
53
.
More specifically, when a pixel having a pixel value of “1” is detected during the scanning of the image
20
in a first scanning operation, it is decided whether or not there are any adjacent pixels having a label number given thereto. If there is an adjacent pixel having a label given thereto, the same label number as in the adjacent pixel is given to the detected pixel. When a plurality of adjacent pixels have the label numbers given thereto, the same label number as in the adjacent pixel having a higher priority is given based on a predetermined priority. When there is no adjacent pixel having the label number given thereto, a new label number is given to the detected pixel. Thus, the inside of the image is scanned to give the label number of the pixel having the pixel number of “1”.
As described above, in the image having the label number given thereto, pixels having the same label number given thereto form one connecting region. However, a plurality of label numbers are given to one connecting region depending on the shape of the connecting region in some cases. As shown in
FIG. 21
, for example, a connecting region A of the image
20
is labeled as three regions having label numbers
2
,
3
and
4
in an image
20
′ as a result of the above-mentioned labeling. In the labeling process, a link table
51
is created. The link table
51
stores information indicating that three regions having the label numbers
2
,
3
and
4
are identical. Furthermore, label editing is carried out to create a conversion table
53
for setting the label numbers
2
,
3
and
4
to
2
and re-labeling the label numbers
3
and
4
thus generating a vacancy into a region having label numbers
5
and
6
.
Then, in a second scanning operation, each of the pixels of the image
20
′ is scanned and a label number is converted (re-labeled) into a real label number by referring to the information of the conversion table
53
. Consequently, each region is finally labeled correctly as in an image
20
″.
In the above-mentioned method, however, it is necessary to scan the whole image twice. Therefore, a long time is taken to perform the labeling process. The time required for the processing is prolonged as the size of the image is increased.
SUMMARY OF THE INVENTION
The present invention is directed to solve the above-mentioned problem, and has an object to provide an image processing apparatus and method capable of retrieving a specific pattern from an image at a high speed and a recording medium storing a program therefor.
The invention has another object to provide an image processing apparatus and method for a labeling processing in the connecting region of an image at a high speed, and a recording medium storing a program therefor.
In a first aspect of the invention, an image processing apparatus is provided for retrieving a specific pattern of an image. The apparatus comprises a unit for dividing the image into a plurality of block regions, each of the block regions having a predetermined size, a unit for creating a map image comprising a plurality of pixels, each of the pixels corresponding to one of the block regions by one to one, a unit for extracting features of the image for each of the block regions, a unit for deciding whether or not the extracted feature meets a predetermined condition, a unit for labeling the pixel on the map image based on a result of the decision, the pixel corresponding to the block region of which feature meets the condition, a unit for detecting a predetermined pattern for specifying a search region on the image from the labeled map image, a unit for specifying the search region based on a position on the map image in which the predetermined pattern is detected, and a unit for retrieving the specific pattern in the specified search region on the image.
In the apparatus, at least one of a concentration value of a pixel, an edge amount and a co-occurrence characteristic may be extracted as the feature amount. In the apparatus, the predetermined pattern for determining the search region may be detected by using a filter having the same size as a size of the specific pattern in the map image.
In a second aspect of the invention, an image processing method is provided for retrieving a specific pattern of an image. The method comprises dividing the image into a plurality of block regions, each of the block regions having a predetermined size, creating a map image comprising a plurality of pixels, each of the pixels corresponding to one of the block regions by one to one, extracting features of
Hashimoto Keisuke
Murakawa Akira
Mariam Daniel G.
Minolta Co. , Ltd.
LandOfFree
Image processing apparatus and method for recognizing... 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 processing apparatus and method for recognizing..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Image processing apparatus and method for recognizing... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3179283