Abstract: A method and apparatus for improving an OCR process for recognizing unidentified characters through the creation of sets of like unidentified characters in a scanned image, deducing what the unidentified characters are with a spell checking procedure and replacing only those unidentified characters which are unambiguously determined. The order for identifying characters is strategically performed so the unidentified characters which are easier to recognize are processed before those that are more difficult to recognize. As characters become identified, relationships defined by corresponding sets allow replacing the same character in related but different textual components. As characters become recognized, other characters become more recognizable by virtue of belonging to the same textual component, such as a word, or to a related textual component such as another word containing the same character.