Luster number estimation and skeleton refining algorithms for Arabic characters

Abstract EN

-This paper presents two techniques for estimating the number 0؛ clusters for automating a clustering based skeletonization algorithm (CBSA) for Arabic characters.

The analysis indicates that an approximate number of clusters suffices for this type of application.

The cluster number estimation algorithms are based on estimating the approximate skeleton length of the character.

This length is either used directly or standardized first.

Experimental results show that the direct method is, in general, better.

Using the above techniques may produce skeletons with redundant loops.

Skeleton refining algorithms are introduced to eliminate any redundant loops.

The experimental results show that the combination of the cluster number estimation, the CBSA skeletonization algorithm, and the refining algorithms produce acceptable skeletons that do not require any a priori knowledge of the data.

These skeletons will prove to be an essential step in modeling and in Arabic optical character recognition techniques using this approach.