Recognition of Chinese Characters by Growing Stroke Window

Abstract

In this paper, we propose a novel Chinese character recognition method. In our approach, the 2-D input character is first converted to a 1-D stroke string. After stroke extraction, the stroke sequence is rearranged in which strokes with similar orientation are located consecutively. By special organizing, strokes that fail to be extracted correctly will be located adjacently in the rearranging stroke string. Some intra-character spatial relationships are recorded in an adjacently matrix. In the recognition stage, the adjacency matrix of input character and that of character in the database are compared. Since both the two adjacency matrices are built by its own character, the relative relationships between strokes will not be influenced by rotation and translation distortions. Finally, the similarity between input character and reference character show that the proposed method is feasible. Besides, it can also resolve the problems inherited from the stroke extraction stage.