The paper presents an algorithm for the estimation of skeletons of thick characters. We directly identify the core pixels of the skeleton forming the core skeletal segments based on labeling of the character boundary with some local properties. The core skeletal pixel is defined as the midpoint of a line segment normal to the boundary pixels. These core skeletal segments are extended and joined systematically based on certain global properties resulting in the final skeleton. The algorithm is independent of the width of the character and is capable of yielding a skeleton close to our intuitive notion of character shape. The topological description of the character is constructed more or less as a by-product of the skeletomization process. The description forms the basis for character recognition using syntactic methods. The algorithm is well suited for parallel implementation.
[1]
C. J. Hilditch.
Comparison of thinning algorithms on a parallel processor
,
1983,
Image Vis. Comput..
[2]
R. S. Ledley,et al.
Automatic pattern recognition for clinical medicine
,
1969
.
[3]
A. G. Sartori-Angus,et al.
Fast thinning algorithm for binary images
,
1985,
Image Vis. Comput..
[4]
R. Mahesh K. Sinha.
PLANG - A picture language schema for a class of pictures
,
1983,
Pattern Recognit..
[5]
R. Mahesh K. Sinha,et al.
Comments on 'fast thinning algorithm for binary images'
,
1986,
Image Vis. Comput..
[6]
J. Sklansky,et al.
Skeleton generation fromx, y boundary sequences
,
1981
.
[7]
E. R. Davies,et al.
Thinning algorithms: A critique and a new methodology
,
1981,
Pattern Recognit..