Thinning Algorithm Research on Pattern Reconstruction and Shape Preservation

• Main Authors: Ming-Te Chao, 趙明德
• Other Authors: Yung-Sheng Chen
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/97uv5a

博士 === 元智大學 === 電機工程學系 === 106 === Thinning algorithm has played an important role in digital image processing. Pattern reconstruction and shape preservation are the even more fundamental requirements of thinning algorithm. Both the widely used rule-based parallel thinning and distance-based medial axis transform have suffered from skeleton distortion. In a rule-based parallel thinning process, the increased iteration count caused by the incessant accumulation of the hidden deletable points (HDP) may give rise to skeleton distortion. Besides, the fork skeleton is split into more fork skeletons. Similarly, the distance-based medial axis transform cannot make reference to the angular boundary corner points and thus the derived skeleton is unable to reflect the angular shape of the pattern. When it comes to pattern reconstruction, although the existing reconstructable parallel thinning is able to realize complete pattern reconstruction with the strategy of embedding the skeletal points extracted by morphological skeleton transform into thinned skeleton, the derived skeleton is seriously disturbed by noisy branches. Besides, with the disk-reconstruction scheme, the distance-based medial axis transform is unable to reconstruct the curved portions of a pattern completely. In view of the above-mentioned, the thesis proposed shape preserving method (RHDP) and pattern reconstructing mechanism (RSP) for rule-based parallel thinning. Furthermore, a new distance-based MAT thinning algorithm is proposed to increase the boundary noise immunity for thinning algorithm. Based on this algorithm, a novel cross-section line based shape descriptor is designed to meet the requirements of complete pattern reconstruction and shape preservation for the thinning algorithm research.