Image analysis – Pattern recognition – Feature extraction
Reexamination Certificate
1997-07-15
2002-10-08
Chang, Jon (Department: 2623)
Image analysis
Pattern recognition
Feature extraction
C382S217000, C382S243000
Reexamination Certificate
active
06463176
ABSTRACT:
BACKGROUND OF THE INVENTION
This invention relates to a method and apparatus for recognizing objects and for reproducing the images of objects.
DESCRIPTION OF THE RELATED ART
One known example of a conventional method of displaying figure patterns or the like is a technique used in inputting figures as disclosed in the specification of Japanese Patent Publication (KOKOKU) No. 5-36830. The technique includes expressing the bent portion of the stroke of a geometrical figure symbol by one of various bend patterns (shape primitives) prepared in advance and approximating curves by circular arcs. The technique is applicable to recognition of geometrically simple figures.
Further, an example of a method of recognizing objects is employed in an object recognition apparatus disclosed in the specification of Japanese Patent Publication (KOKOKU) No. 5-23463. This method involves tracing the contour of a recognized object, dividing up the contour into shape primitives such as straight lines or circular arcs, registering the attribute of each shape primitive and the attribute of each vertex in memory in the form of a dictionary, searching the dictionary in the memory and subjecting an unknown object to recognition by matching the shape primitives thereof with those of the recognized object.
In the example of the prior art described above, the relationship among the shape primitives arranged in space is not extracted for the purpose of recognition. Therefore, in a case where image patterns differ depending upon the point of view, even though the objects are identical, or in a case where a change in shape or size is produced for some reason, it is required that an enormous amount of two-dimensional pattern information be stored with regard to the identical subjects and that matching be carried out relative to this voluminous pattern information at the time of pattern recognition. A problem that arises is an increase in cost of computations.
Further, in a case where a plurality of objects exist in an image, it is required that the image be partitioned beforehand into appropriate areas. After area partitioning is performed in such a manner that one object will exist in one area, recognition processing is applied.
Such area partitioning and recognition of object patterns are two sides of the same thing and it is very difficult to execute them perfectly in automatic fashion.
Further, there are prior-art examples of techniques for reconstructing an object that has been recognized. For example, the specification of Japanese Patent Application Laid-Open (KOKAI) No. 5-12413 discloses an image generating apparatus in which the three-dimensional shapes and distances of objects in an image are taken into account and images corresponding to points of view are generated by just a few items of data. This method involves using a three-dimensional measuring device to previously measure points on a surface on which an object of interest can be reproduced and storing the three-dimensional coordinates in memory. When a reproduced image is generated, positions on a screen for a reproduced image are obtained from the positional relationship between the point of view and the object of interest, and a display is presented in accordance with luminance level or RGB values from the relationship among corresponding points relative to an image obtained previously at a different point of view.
Further, in a synthesized-image generating apparatus disclosed in the specification of Japanese Patent Publication No. 5-60148, a reference image and a three-dimensional model of an object of interest are stored in a prescribed memory in advance. At playback, the zoom rate of the image of interest, distance of movement on a screen and angle of rotation are computed from disparities with respect to distance and azimuth as seen from a point of view different from that at the time the reference image and three-dimensional model were stored, and a geometrical transformation based upon these computed values is applied to generate an image of interest as observed from the different point of view.
With this example of the prior art, a more detailed three-dimensional model relating to the object of interest is required in order to process the image of a specific target in the image of an actual photograph and combine the image with that of a different scene, or to move and transform the specific target while taking masked portions into account.
Further, in order to obtain a more detailed three-dimensional shape model relating to the object of interest, there is a need for detailed measurement as by a technique for obtaining the shape model from a parallax image acquired by shape measuring means or by binocular stereoscopic camera. As a result, computation costs are high and processing requires an extended period of time.
SUMMARY OF THE INVENTION
Accordingly, an object of the present invention is to provide a pattern recognition method and apparatus in which recognition is performed based upon a limited number of feature elements and information relating to the relative disposition of the feature elements in space.
Another object of the present invention is to provide an image generating method and apparatus in which images are generated at low computation cost and in a short processing time without use of a detailed three-dimensional shape model.
According to the present invention, the foregoing objects are attained by the apparatus or methods described below.
Specifically, the present invention provides an image recognition method comprising an extraction step of extracting local feature elements of an image, and a selection step of selecting a pair composed of a prescribed local feature element and position information indicative thereof, this pair being such that the distance between a pair composed of a prescribed local feature element and position information indicative thereof and a pair composed of a local feature element extracted at the extraction step and position information indicative thereof is less than a prescribed distance.
Further, the invention provides an image reproduction method comprising an extraction step of extracting a feature element immediately prior to occurrence of an occlusion in which at least part of a feature element of an object is hidden behind another feature element, a second feature-element generation step of generating a second feature element on the basis of a plurality of feature elements immediately prior to occurrence of the occlusion extracted at the extraction step,; and a step of reproducing an image on the basis of the second feature element generated by the second feature-element generation step.
Further, the present invention provides an image recognition apparatus comprising extracting means for extracting local feature elements of an image, and selecting means for selecting a pair composed of a prescribed local feature element and position information indicative thereof, this pair being such that the distance between a pair composed of a prescribed local feature element and position information indicative thereof and a pair composed of a local feature element extracted by the extracting means and position information indicative thereof is less than a prescribed distance.
Further, the invention provides an image reproduction apparatus comprising reproducing means for reproducing an image on the basis of interpolation between the pair composed of the prescribed local feature element and position information indicative thereof selected by the selecting means and a pair in the vicinity thereof.
Further, the invention provides an image reproduction apparatus comprising extracting means for extracting a feature element immediately prior to occurrence of an occlusion in which at least part of a feature element of an object is hidden behind another feature element, second feature-element generating means for generating a second feature element on the basis of a plurality of feature elements immediately prior to occurrence of the occlusion extracted by the extracting means, and means for reprodu
Iijima Katsumi
Matsugu Masakazu
Canon Kabushiki Kaisha
Chang Jon
Morgan & Finnegan , LLP
LandOfFree
Image recognition/reproduction method and 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 recognition/reproduction method and apparatus, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Image recognition/reproduction method and apparatus will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2974071