Isolated Character Detection

With the advent of powerful, user-friendly graphic arts and page layout tools, historically simple single-font, text-only documents are now easily produced with a montage of color, line graphics, images and multi-font text. These additions greatly enhance the presentation value of every conceivable document type, from corporate annual reports, to advertisements and magazine articles - even mere personal correspondence. Indeed, such multi-component documents are now the de facto standard for nearly every exchange of printed information, not to mention the rapidly growing spectrum of digital documents. To keep pace with increasingly sophisticated document generation techniques, document reproduction technology must also evolve to ensure high quality replication of the text, line graphics, and images found in a wide variety of document types.

Today, conventional optical character detection (OCD) and optical character recognition (OCR) algorithms yield acceptable detection and recognition of characters arranged in columnized, clear-text blocks. These and other algorithms form the basis for state-of-the-art document analysis systems, which are capable of a reasonably accurate partitioning of documents into clear-text, line graphics, and image components. Each component can be processed separately for reproduction, providing high quality output. However, documents often contain isolated characters and small text-segments, in both clear-text form, and embedded within line graphics and images. The problem of reliably detecting such characters and small text-segments, which occur in arbitrary shapes, styles, sizes, and orientations, is of great complexity and is as yet unsolved.

The same fundamental technologies necessary to accomplish isolated character and small text-segment detection are directly applicable to digital document storage and retrieval, in particular to efficient indexing into digital document databases. Text embedded in line graphics and images are natural keys for indexing operations, and can be used to perform an efficient search through stored pictorial information.

Last updated: June 3, 1998

Back to research topics.