Image analysis – Pattern recognition – Feature extraction
Reexamination Certificate
1998-03-05
2001-05-08
Au, Amelia (Department: 2623)
Image analysis
Pattern recognition
Feature extraction
C382S203000
Reexamination Certificate
active
06229920
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention concerns a method for classifying and recognizing patterns and in particular a method whereby a pattern may be classified by utilizing a classification vector which differs depending on a degree of fractality of the respective pattern.
2. Discussion of the Background
Pattern recognition constitutes a central problem in many technical fields. Such pattern recognition should make it possible to acquire m-dimensional objects by using data processing means in such a manner that the data processing means is enabled to determine at maximum possible accuracy to which m-dimensional object the respective pattern should be assigned. Highly precise pattern recognition of m-dimensional objects would, for example, permit automatic navigation of any kind of vehicles to thus prevent accidents brought about by human error as far as possible. It would furthermore be possible to automatically and highly accurately identify the handwriting of any person. In the same manner, production of automatic machines or robots equipped with an intelligent sensory mechanism would not constitute a problem with highly accurate pattern recognition. Further fields of application are e.g. in the recognition of contour lines provided in the form of red-green-blue color data, of a symbol in symbol input via a pressure-sensitive digitising tablet, or of monaural audio data etc. Many other applications are equally conceivable.
Many methods for pattern recognition are already known in the prior art. A drawback of each known method does, however, reside in the fact that they are only applicable to specific kinds of objects. Universal applicability of these known methods is consequently strongly limited. It is another drawback of the known methods that in particular reliable recognition of highly complex structures or fractal structures will either fail or only succeed at extremely high computing speeds, such that real-time pattern recognition will not be possible.
In pattern recognition it is frequently also necessary to disregard certain features of the object or to only consider them in such a way that the actual measure reflecting a feature will become irrelevant. Thus it is conceivable e.g. that the actual spatial extension of an object should be irrelevant to thereby make it possible that objects merely differing in their spatial size are recognized to be similar.
SUMMARY OF THE INVENTION
It is therefore the object of the present invention to furnish a method for classifying and recognizing patterns which is applicable to any type of object and which classifies and recognises patterns on the basis of a classification vector which differs in accordance with the degree of fractality of the object.
Thus object is attained by the measures indicated in the claims.
Further advantageous embodiments of the present invention result from the subclaims.
In particular, in accordance with the method according to the claims a pattern of an m-dimensional object to be classified and recognised and having the form of an m-dimensional traverse is established. For each point of this traverse a property is detected which unambiguously reflects a relationship of a respective point of the traverse with the point preceding it and the point succeeding it. These properties for each point are linked in order to obtain an overall property unambiguously characterising the traverse. Subsequently the traverse is smoothed. After this, the steps of acquiring, linking and smoothing are repeated (k−1) times, with k representing an integer. By using the obtained overall properties present in the number k, a signature is generated for the pattern. This signature is used for comparison with signatures of known patterns in order to ascertain a degree of similarity between the compared signatures.
By means of this method it is possible in a simple manner to classify a pattern of any m-dimensional object on the basis of the degree of fractality of the pattern not only be means of a dimension of fractality but a whole set of numbers, i.e. a vector comprising k components. If the traverse utilized in the method comprises n points, the pattern comprising n points and the dimension m is consequently classified by a vector having k components, thereby enabling more rapid and more data-extensive comparison. As the method is not subject of any dimensional restrictions, it may furthermore be applied to any desired objects. By making appropriate use of various conditions or demands in acquisition of the properties and smoothing of the traverse, the method may be modified such as to disregard certain characteristics of the pattern to be classified and recognized, or for instance only considers them in such a manner that differences of scale between the pattern to be classified and recognized and the pattern of a known object are not taken into account. This means that the method may be modified to adapt to any conceivable application.
REFERENCES:
patent: 3942169 (1976-03-01), Fujimoto et al.
patent: 4317109 (1982-02-01), Odaka et al.
patent: 4365235 (1982-12-01), Greanias et al.
patent: 4542412 (1985-09-01), Fuse
patent: 4653107 (1987-03-01), Shijoma et al.
patent: 4703363 (1987-10-01), Kitamura
patent: 4748675 (1988-05-01), Suzuki et al.
patent: 4760605 (1988-07-01), David et al.
patent: 4791679 (1988-12-01), Barski et al.
patent: 5036544 (1991-07-01), Sakaue et al.
patent: 5050227 (1991-09-01), Furusawa et al.
patent: 5150431 (1992-09-01), Yoshida et al.
patent: 5218649 (1993-06-01), Kundu et al.
patent: 5345547 (1994-09-01), Anezaki et al.
patent: 5448653 (1995-09-01), Hori et al.
patent: 5598215 (1997-01-01), Watanabe
patent: 5617486 (1997-04-01), Chow et al.
patent: 5706419 (1998-01-01), Matsugu et al.
patent: 5740273 (1998-04-01), Parthasarathy et al.
Kass, M., Witkin, A., Terzopoulos, D., Snakes: Avtive Contour Models, International Journal of Computer Vision, pp. 321-331.*
Jia-Guu Leu, Pattern Recognition, vol. 24, No. 10, pp. 949-957, “Computing A Shape's Moments From Its Boundary”, Jan. 1, 1991.
Au Amelia
Delphi Systemsimulation GmbH
Miller Martin
Oblon & Spivak, McClelland, Maier & Neustadt P.C.
LandOfFree
Method of classifying and recognizing patterns by additive... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method of classifying and recognizing patterns by additive..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method of classifying and recognizing patterns by additive... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2449753