A Character Extraction and Recognition Method for Line Drawings

Character segmentation and recognition are two difficult key steps in extracting literal information from images. This paper proposes a method for effective segmentation and recognition of characters in complicated graphical contexts of line drawings. Text pixels are first separated from intersecting non-text objects with the holistic graphic recognition algorithm, and then character boxes are grouped into strings using various contextual constraints. The individual characters of a string are recognized with an adaptive recognition-based segmentation scheme based on the structural stroke characteristics. The efficiency and stability of the proposed method have been shown by the experiment results on practical drawings.