Authors: Paul Corey and Tracy Hammond
Summary:
This paper proposes a method to integrate a feature based classifier Rubine and a geometric recognition system based on LADDER. Rubine classifier is used to calculate minimum mahalanobis distance of the sample from the classes. Then the sample is rejected for the rubine classification based on a threshold. This threshold value is set to 35 which is claimed to be optimum threshold.
The sample which falls above the threshold are classified using geometric recognition system.
Discussion:
Its kind of clever way to improve accuracy by using geometric recognition system to classify samples that may be rejected by Rubine. I expected a significant increase in the accuracy of classification. As the paper states, the shapes which fall in the common area of both these systems may be the reason for errors(reduced accuracy).
Subscribe to:
Post Comments (Atom)
1 comment:
If you can't make a better approach, then combine approaches. Works for me. Now if we can incorporate Olsen's thesis, we'll have a triple threat.
Post a Comment