Text Extraction Algorithms for Document Images Including Forms
Jia Wang · Shuju caiji yu chuli · 2004
A text extraction algorithm is proposed for document images including forms in the page. The foreground pixels are extracted with bounding boxes by mask scanning, then, pattern lists are formed by grouping the neighbored boxes according to a certain criteria. Features such as the maximum black run length, the height, and the width of the patterns are employed for the pattern classification. Experimental results show that the proposed algorithm can successfully tackle normal forms and extract text from skewed forms and the flow chart. The algorithm is valid only for binary document images.