Proposed heuristic procedures to preprocess character patterns using line adjacency graphs

摘要

Heuristic procedures are proposed to remove noise from character patterns. The procedures use line adjacency graphs (LAGs) of two types: white and black. Our proposed white LAGs are an extension of the well-known black LAGs. Furthermore, we refine the procedure to condense black LAGs into different types of nodes, the types of node depending on the local configuration of the pattern at that node. Finally, specialized procedures are developed to skeletonize the different types of nodes, which in turn skeletonize the full pattern. Experiments demonstrate that our noise removal and skeletonization procedures are robust and fast. Our results compare favorably with those of established algorithms.