A document segmentation, classification and recognition system

Abstract

This paper proposes a document segmentation, classification and recognition system for automatically reading daily-received ofice documents that have complex layout structures, such as multiple columns and mixed-mode contents of texts, graphics and half-tone pictures. First, the block segmentation employs a twostep run-length smoothing algorithm for decomposing any document into single-mode blocks. Next, based on cluslering rules the block classification classifies each block into one of text, horizonld or vertical lines, graphics, and pictures. The text block is separated into isolated characters using projection profiles, and which are translated into ASCII codes through a font-and sizeindependent character recognition subsystem. Logo pictures discriminated from half-tone pictures are identified and converted into symbolic words. The experimental results show that the proposed system is capable of correctly reading different styles of mixed-mode printed documents.

Research of Artificial Neural Networks Abilities in Printed Words Recognition

“…Approaches used to tackle this problem can be divided into: top-down and bottomup. The most common top-down techniques are run-length smoothing [20]- [21] and projection profiles [22]- [25]. Topdown approaches are splitting image into regions which are later identified and further spitted into text columns then paragraphs, text lines and finally words [23].…”

Document image segmentation using fuzzy classifier and the dual-tree DWT

“…Top-down approaches split the document into blocks, columns paragraphs, text lines, or even words. The most common top-down techniques are run-length smoothing [2], [3] and projection profiles [4]. These methods are not suitable for skewed texts because top-down methods are restricted only to rectangular blocks.…”

Wavelet‐based images compression of color document by fuzzy picture‐text segmentation

“…Tsai (1985) proposed an approach to automatic threshold selection. Some other systems based on the prior knowledge of some statistical properties of various blocks (Fisher et al, 1990;Akiyama and Hagita, 1990;Shih et al, 1992;Pavlidis and Zhou, 1992;Zlatopolsky, 1994), or texture analyses (Wang and Srihari, 1989;Jain and Bhattacharjee, 1992) have also been developed. Suen and Wang (1996) presented a text string extraction algorithm, which uses the edgedetection technique and text block identification to extract text strings.…”