Text Block Recognition in Multi-Oriented Handwritten Documents
Rainer Herzog, Arved Solth, Bernd Neumann · 2014
Abstract—Automatic detection of text blocks is an important step before applying OCR or word-spotting techniques to docu-ment images. Our approach focusses on handwritten (historical) documents and uses the Gabor Transformation to facilitate this task. Apart from the main text, which often consists of rectangular shaped text blocks, marginalia are of special interest here. These areas are generally unconstrained regarding size, dimensions or orientation. Our system detects text blocks of at least three lines, representing a moderately homogeneous region regarding orientation and distances of text lines. Experiments on 40 documents, written in different european and asian writing systems, show good results, depending on the complexity of the layout. Keywords-document layout analysis; manuscript; text block recognition; Gabor Transform I.