Automatic separation of text from background in scanned images of complex documents

摘要

A system that converts a scanned image of a complex document into an image where text has been preserved and separated from the background. The system first subdivides the scanned image into blocks and then examines each block pixel by pixel to construct a histogram of the gray scale values of the pixels. The histogram is partitioned into a first, middle and last regions. If one or more peaks occur in the first and last regions, and a single histogram peak occurs within the middle region, the pixels are reexamined to determine the frequency of occurrence of pixels having a gray scale level of the middle peak nearby pixels which have a level of a first region peak. If this frequency is high, the middle peak is assumed to be background information. After determining the threshold, the system rescans the block applying the threshold to separate the text from background information within the block.