Shape Decomposition and Stroke Extraction of Handwritten Korean Characters
Jeong-Sun Park, Kicheon Hong, Il-Seok Oh · International Journal of Computer Processing Of Languages · 2002
For reliable recognition of the handwritten Korean characters, stroke extraction is one of the most important tasks. Classical approaches are based on the thinning operation and they have a limitation due to the problem of distorting original pattern shape. In this paper, we propose a shape decomposition algorithm that decomposes a handwritten Korean character pattern into a set of near-convex parts. The algorithm is designed based on the observation that Korean character patterns can be decomposed using two shape features called T-junction and B-junction. To handle robustly complex situations in which three or more strokes meet at a joint, we detect joint parts from the shape-decomposed pattern. We classify the parts into several types depending on their interrelationships. A line segment is extracted from a part such that it best represents the part and keeps connectivity with neighboring parts. We develop five criteria for the quality evaluation of our method and the thinning-based methods. The evaluation shows that our method is superior to the thinning methods and our results are very promising for the subsegment processes like recognition.