A robust parallel thinning algorithm for pattern recognition

Abstract

Thinning algorithms have been used in pattern recognition and image analysis for a long time. They reduce a binary digital pattern to obtain a unit width skeleton which retains geometrical and topological properties. These properties are important for robust recognition of characters, handwritings, fingerprints, transportations infrastructure and other. A robust parallel thinning algorithm based on the popular Zhang and Suen algorithm is presented. The ZS algorithm is very good in respect to both connectivity and insensitivity to boundary noise but it tends to remove diagonal line segments and whole 2×2 square patterns, and does not produce a unit width skeleton. The experimental results show that the proposed method preserves good properties of ZS algorithm and it overcomes the disadvantages by incorporating additional conditions for identifying the crucial patterns and by applying a post-processing step that removes all redundant pixels so the one pixel thick skeleton is produced.