Classification system with reject class

Details Classification system with reject class Classification system with reject class

2000-09-05

2004-02-10

Mehta, Bhavesh M. (Department: 2625)

Image analysis

Pattern recognition

Classification

C382S156000, C382S160000, C382S228000, C706S013000, C706S020000, C706S026000

Reexamination Certificate

active

06690829

ABSTRACT:

BACKGROUND OF THE INVENTION
The invention concerns a method of signal processing, wherein signals S are checked for belonging to objects of desired classes Z
k
and differentiated from objects in an undesired class Z
A
.
DESCRIPTION OF THE RELATED ART
In order not to skip over any objects of relevance, segmenting algorithms produce, in general, a large number of hypotheses whose examination requires a large expenditure of time. Another disadvantage is that a segmenting algorithm can often only consider a small number of attributes of the object to be segmented, such as shape or color, in order to analyze in real time a complete picture or at least a region of a picture in which new objects can knowingly emerge. In the case of a segmentation of circular traffic signs in a street scene, the segmentation is performed, in general, via a Hough-transformation (K. R. Castelman; Digital Image Processing, Prentice Hall, New Jersey, 1996) or a distance transformation that is based upon a matching algorithm (D. M. Gavrilla; Multi-Feature-Hierarchical Template Matching Using Distance Transforms, IEEE Int. Conf. on Pattern Recognition, pp 439-444, Brisbane, 1998), all in order to find all shapes typical for a circular traffic sign.
For a classification, which follows such a segmentation, the principal problem is not the differentiation of the different objects of the class Z
k
(e.g. traffic signs) from each other, but rather the difficulty of differentiating the objects of this class Z
k
from the objects of the undesired class Z
A
. Therein the objects of the undesired class Z
A
consist of any image field which are selected by the segmentation algorithm due to their similarity to objects of the class Z
k
.
This leads to a two class problem, in which certainly only the class Z
k
is more or less locatable in the feature realm or set, while the class Z
A
is widely scattered over the feature set. Therein, in general, it is not possible to find a limited number of ‘typical’ objects, that are associated with the class Z
A
. Were number of objects of the class Z
A
so limited, then it would be possible to generalize a classifier starting from a set of learning examples of the total variations of possible elements of the class Z
A
; assuming that objects are limited to those from a closed world (closed world assumption) from which the general classification theory originates is damaged in this case.
In reality, most of the objects produced by a segmentation algorithm belong to the class Z
A
. In the case of traffic sign recognition these are typically more than 95 percent, which further complicates the classification problem.
SUMMARY OF THE INVENTION
The task of the invention is to develop a signal processing method, which has high certainty while processing the signal in real time and avoiding a false classification of objects of the class Z
A
, i.e. that the probability is kept low that objects of class Z
A
will be falsely assigned to one of class Z
k
.


REFERENCES:
patent: 5123057 (1992-06-01), Verly et al.
patent: 5966701 (1999-10-01), Kohda et al.
patent: 6337927 (2002-01-01), Elad et al.
patent: 6345119 (2002-02-01), Hotta et al.
patent: 6549661 (2003-04-01), Mitsuyama et al.
patent: 44 04 775 (1995-07-01), None
patent: 198 02 261 (1999-07-01), None
Aksela, M.; Laaksonen, J.; Oja E; Kangas, J.; Document Analsyis and Recongnition, 2001. Proceedings. Sixth International on, Sep. 10-13, 2001 pp. 982-986.*
Niemann, H., “Klassifikation von Mustern”, Spinger-Verlag, Berlin Heidelberg New York Tokyo 1983.
Schurmann, J., “Palfern Classification”, Chapter 10, Reject Criteria and Classifier Performance, 1996, Wiley-Intersciences, New York.
Oren, Michael, et al., “Pedestrian Detection Using Wavelet Templates”, IEEE Conf. on Computer Vision and Pattern Recognition, S. 193-199, San Juan, 1997.

Affiliated with

Also associated with

Rating

Classification system with reject class does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Classification system with reject class, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Classification system with reject class will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-3340173