Apparatus and method for image recognition

• US 20040062435A1
• Filed: 10/02/2003
• Published: 04/01/2004
• Est. Priority Date: 09/30/1999
• Status: Active Grant

• Current Assignee: Kenji Nagao, Megumi Yamaoka
• Original Assignee: Kenji Nagao, Megumi Yamaoka
• Inventors: Yamaoka, Megumi, Nagao, Kenji
• Granted Patent: US 7,054,489 B2
• Time in Patent Office: Days
• US Class Current: 382/159
• CPC Class Codes: G06V 10/443 by matching or filtering

Abstract

An object recognizing apparatus is provided which is capable of precisely recognizing an object in an input image with the use of a corresponding learning image even when a local-segment of the input image coincides with a learning-local-segment of another similar learning image. The apparatus comprises (1) image dividing means for dividing an input image, which is received from image input means, into local-segments, (2) similar-local-segment extracting means for extracting a similar learning-local-segment to the local-segment of the input image from a learning image database, (3) object position estimating means for estimating the position of an object to be identified in the input image from the coordinates of the local-segment and the coordinates of the learning-local-segment corresponding to the local-segment, (4) counting means for counting the local-segments coincide with their corresponding learning-local-segments, and (5) object determining means for judging that the object to be identified is present when a result of counting is greater than a predetermined number. Consequently, the object and its position in any input image can be detected at higher accuracy.

Claims

1. An image recognizing method comprising the steps of:

(a) dividing an input image into local-segments;

(b) registering a learning image into a learning image database;

(c) extracting a learning-local-segment which is similar to one of the local-segments from the learning image database;

(d) relating the learning-local segment extracted in the step (c) to the one of the local-segments;

(e) estimating a position of an object to be identified in the input image from coordinates of the one of the local-segments and coordinates of the learning-local-segment;

(f) counting a pair of one of the local-segments and learning-local-segment from which a first position is estimated to determine a score for the first position; and

(g) judging that the object to be identified is present at the first position when the score is greater than a predetermined number.

3. The image recognizing method according to claim 1, wherein:

said step (b) comprises the step of registering the learning image into the learning image database by a character of an object to be identified;

said step (c) comprises the step of extracting the learning-local-segment which is similar to the one of the local-segment from the learning image database by the character; and

said step (f) comprises the step of counting a pair of one of the local-segments and a learning-local-segment by the character.

5. The image recognizing method according to claim 1, wherein the step (d) comprises the steps of:

(d-1) calculating a sum of one of (i) each square of a difference between a pixel value of the one of the local-segment and a pixel value of the learning-local-segments and (ii) each absolute of the difference, and extracting a pair of one of the local-segments and a learning-local-segment which has minimum one of the sum; and

(d-2) relating the one of the local-segment to the earning-local-segment in the pair extracted in said step (d-1).

2. An image recognizing method comprising the steps of:

(a) dividing an input image into local-segments;

(b) dividing a learning image into learning-local-segments having a same size as the local-segments and making a group of some of the learning-local-segments which are similar to each other;

(c) registering image data of a representative earning-local-segment of the group and coordinates of all the some of the learning-local-segments into a same-type window database;

(d) extracting a representative learning-local-segment which is similar to one of the local-segments from the same-type window database;

(e) relating the one of the local-segments to a group of which the representative learning-local-segment extracted in the step (d);

(f) estimating a position of an object to be identified in the input image from coordinates of the one of the local-segment and coordinates of the representative learning-local-segment of the group;

(g) counting a pair of one of the local segments and a representative learning-local-segment from which a first position is estimated to determine a score for the first position; and

(h) judging that the object to be identified is present at the first position when the score is greater than a predetermined number.

4. The image recognizing method according to claim 2, wherein said step (c) comprises the step of registering image data of the representative learning-local-segment of the group and coordinates of all the some of the learning-local-segments in the group and a character of an object to be identified into the same-type window database.

6. The image recognizing method according to claim 2, wherein said step (e) comprises the steps of:

(e-1) calculating a sum of one of (i) each square of a difference between a pixel value of the one of the local-segment and a pixel value of the representative learning-local-segment and (ii) each absolute of the difference, and extracting a pair of one of the local-segment and a representative learning-local-segment which has minimum one of the sum; and

(e-2) relating the one of the local-segment to the representative learning-local-segment in the pair extracted in said step (e-1).

7. An image recognizing apparatus comprising:

image dividing means for dividing an input image into local-segments;

learning means for registering a learning image into a learning image database;

similar window extracting means for extracting a learning-local-segment which is similar to one of the local-segments from the learning image database and for relating the learning-local-segment to the one of the local-segment;

object position estimating means for estimating a position of an object to be identified in the input image from coordinates of the one of the local-segment and coordinates of the learning-local-segment;

counting means for counting a pair of one of the local-segments and a learning-local-segment from which a first position is estimated by said object position estimating means to determine a score for the first position to determine a score for the first position; and

object determining means for judging that the object to be identified is present in the first position when the score is greater than a predetermined number.

8. An image recognizing apparatus comprising:

image dividing means for dividing an input image into local-segments;

learning means for dividing a learning image into learning-local-segments having a same size as the local-segments and for making a group of some of the learning-local-segments which are similar to each other and for registering a representative learning-local-segment of the group and coordinates of all the some of the learning-local segments into a same-type window database;

similar window extracting means for extracting from the same-type window database the representative learning-local-segment of the group which is similar to one of the local-segments of the input image and for relating the learning-local-segments to the one of the local-segment;

object position estimating means for estimating a position of an object to be identified in the input image from coordinates of the one of the local-segment and coordinates of the learning-local-segment;

counting means for counting a pair of one of the local-segments and a learning-local-segments from which a first position is estimated by said object position estimating means to determine a score for the first position; and

object determining means for judging that the object to be identified is present at the first position when the score is greater than a predetermined number.

10. The image recognizing apparatus according to claim 8, wherein said learning means includes:

similar window integrating means for making a group of some of the learning-local-segments which are similar to each other stored in the learning image database and for releasing image data of a representative learning-local-segment of the group and coordinates of all the some of the learning-local-segments in the group; and

a same-type window database for storing the image data of the representative learning-local-segment of the group and the coordinates of all the some of the learning-local-segments in the group.

9. An image recognizing apparatus comprising:

image dividing means for dividing an input image into local-segments;

learning means for registering learning images by a character of a object to be identified into a learning image database;

similar window extracting means for extracting a learning-local-segment which is similar to one of the local-segments from the learning image database by the character and for relating the learning-local-segment to the one of the local-segment by the character;

object position estimating means for estimating a position of an object to be identified from coordinates of the one of the local-segment and coordinates of the learning-local-segment by the character;

counting means for counting a pair of one of the local-segments and a learning-local-segment from which a first position is estimated by said object position estimating means to determine a score for the first position by the character; and

object determining means for judging that the object to be identified is present at the first position when the score is greater than a predetermined number.

11. A computer-readable storage medium holding a program for making a computer carry out an image recognizing method, said image recognizing method comprising the steps of:

(a) dividing an input image into local-segments;

(b) registering a learning image into a learning image database;

(c) extracting a learning-local-segment which is similar to one of the local-segment of the input image from the learning image database;

(d) relating the learning-local-segment extracted in the step (c) to the one of the local-segments;

(e) estimating a position of an object to be identified in the input image from coordinates of the one of the local-segments and coordinates of the learning-local-segment;

(f) counting a pair of one of the local-segments and a learning-local-segment from which a first position is estimated to determine a score for the first position; and

(g) judging that the object to be identified is present at the first position when the score is greater than a predetermined number.

12. An image recognizing apparatus for detecting a shape of an object from an image, comprising:

an image database into which a shape identifier specifying the shape of the object and a model image, which is a image of the object having the shape, are preliminarily registered;

model generating means for extracting characteristic data of the shape from the model image;

a shape database for storing the characteristic of the shape with the shape identifier in a combination;

an image input unit for supplying an input image;

an image cutout unit for cutting out an image segment from the input image;

shape classifying means for comparing the image segment with the characteristic data of the shape to determine whether or not the object of the shape is present in the image segment; and

an output unit for releasing data about the shape of the object determined by said shape classifying means and data about a position of the shape of the object in the input image.

13. The image recognizing apparatus according to claim 12, wherein said model generating means is operative to:

extract an average image of the model image of the shape and a variance of each pixel in the model image as the characteristic data of the shape; and

release a combination of the average image, the variance, and the shape identifier into the shape database.

14. An image recognizing apparatus for detecting a shape of an object from an image, comprising:

an image database preliminarily storing a shape identifier specifying the shape of the object and a model image which is an image of the object of the shape;

model generating means for calculating a base vector in a characteristic space from a pixel value of the model image, for projecting the model image in the characteristic space as a model image vector, for calculating a characteristic statistic value of the shape from the model image vector having the shape identifier as a characteristic shape parameter, and for adding the shape identifier to the characteristic shape parameter;

a shape database for storing the base vector, the characteristic shape parameter, and the shape identifier in a combination;

an image input unit for supplying an input image;

an image cutout unit for cutting out an image segment from the input image;

shape classifying means for projecting the image segment in the characteristic space to determine an image segment vector based on the base vector and for comparing the image segment vector with the model image using the characteristic shape parameter to determine whether or not the shape of the object is present in the image segment; and

an output unit for releasing data about the shape of the object and data about a position of the shape of the object in the input image when an object of which shape coincides the shape to be detected is present in the input image.

15. The image recognizing apparatus according to claim 14, wherein said model generating means is operative to calculate the characteristic shape parameter from an average vector and a covariance of the model image vector derived from the model image.

16. The image recognizing apparatus according to claim 14, wherein said model generating means is operative to calculate an average image of the model image, calculate a base vector from a pixel value of the average image, project the model image in the characteristic space as a model image vector, and add the shape identifier to the model image vector.

17. The image recognizing apparatus according to claim 14, wherein the shape identifier includes data indicating what portion of the object the shape is.

18. The image recognizing apparatus according to claim 17, wherein said shape classifying means is operative to estimate an overall area which the object occupies in the input image from the image segment of the shape identifier and sum up the overall area estimated for the image segment to output a position of the overall area of the object.

19. An image recognizing method for detecting a shape of an object from an image, comprising the steps of:

registering a shape identifier specifying the shape of the object to be identified and an image of the object having the shape as a model image into an image database;

extracting a characteristic data of the shape from the model image;

releasing the characteristic data of the shape and the shape identifier in a combination into a shape database;

supplying an input image to be determined whether or not the object is present therein;

cutting out an image segment from the input image;

comparing the image segment with the characteristic data of the shape to determine whether or not the object of the shape to be identified is present in the image segment; and

releasing data about the shape of the object and data about a position of the shape of the object in the input image.