Handwritten character-based parallel thinning algorithms : a comparative study

Abstract EN

Accurate, efficient and reliable thinning of character patterns is essential to a variety of applications in the field of document analysis and recognition systems.

Several algorithms for thinning of digital patterns, which are frequently cited and proven to produce good results, are compared.

A good thinning algorithm is one that posses the following set of desirable features: maintaining connectivity of resulting skeletons; producing skeletons of unit width; insensitive to noise; generating skeletons around the medial axis; and time-efficient.

We propose to use these measures as a common ground for evaluating the selected thinning algorithms and we limit their applications to handwritten character recognition because of its importance.

The aim of this work is to show how thinning could affect the success rates of recognition systems.

Further, the analysis provides us with insight to why an algorithm performs poorly when compared to another one that performs well on the same pattern? As a result, this study may help to compare, to select, or to improve on any of the selected algorithms or may be to develop a new one appropriate to this particular application.

The comparative results are included to support our findings.