Locating text in color document images

A novel text extraction algorithm from cluttered color document images is developed and tested. The algorithm consists of a color segmentation stage followed by rule-based filtering of non-text regions. Extraction of text segments algorithm uses the measurement of geometrical properties as well as characterness properties and a set of heuristic rules. The algorithm includes a fusion cycle of three different segmentation maps, and a restitution cycle to restore any deleted characters and/or their diacritical marks. The proposed method, proven successful in extraction of texts from many color document images, has applications in color image indexing and retrieval.