Caption and photo extraction from scanned document images

摘要

A bitmap image data (20) is analyzed by connecting component extraction (24) to identify components or connected components that represent either individual characters or letters, or regions of a nontext image. The connected components are classified as text or nontext based on geometric attributes such as the number of holes (60), arcs (50, 52) and line ends (54, 56) comprising each component. A nearest-neighbor analysis (30) then identifies which text components represent lines or strings of text and each line or string is further analyzed (34) to determine its vertical or horizontal orientation. Thereafter, separate vertical and horizontal font height filters (36) are used to identify those text strings that are the most likely candidates. For the most likely title candidates a bounding box (40) is defined which can be associated with or overlaid upon the original bitmap data to select the title region for further processing or display. Captions and photographs can also be located. <IMAGE>