A Robust Table Registration Method for Batch Table OCR
Processing
Jinyu Zuo
Polar Rain, Inc.
900 E. Hamilton, Suite 100
Campbell, CA, 95008
jinyu.zuo@polarrain.com
Esin Darici
Polar Rain, Inc.
900 E. Hamilton, Suite 100
Campbell, CA, 95008
esin.darici@polarrain.com
ABSTRACT
A robust table registration method is proposed in this p aper
for a better understanding on structured information from
scanned table images. Scanned images can be heavily de-
graded because of scanning effects, binarization or purely
docu ment itself. For batch processing images with the same
table structure, normally the table model is provided and
can be used to overcome most challenging quality factors.
The given table model is used as the ground truth in this
paper. H owever, only rough precision is needed on table cell
dimensions and this makes providing the table model an eas-
ier task. The metho d was tested on Multilingual Automatic
Docu ment Classification Analysis and Translation (MAD-
CAT) images and a promising performance is achieved.
Keywords
Table registration, MADCAT, document processing
1. INTRODUCTION
While digitizing existing documents using Optical Char-
acter Recognition (OCR) technology is already a challeng-
ing task, putting table contents back into the original table
structure correctly is another active research topic. There
are many receipts, transcripts, library index, or personal in-
formation cards of employees need to be digitized. However,
how to put the recognized handwritten information back to
the proper field, such as putting the name recognized from
the name field back to the name file in t he XML file, is the
topic that will be discussed in this paper.
Surveys of table processing techniq ues and algorithms can
be found in [4] and [2]. Most techniques in those surveys are
for table detection or table structure analysis instead of reg-
istration. The most similar work can be found in literatures
is [3] where an unsupervised table registration algorithm was
proposed in order to recognize tabular contents. However,
their data were limited to census records.
For most normal tables, the table structure is defined by a
set of horizontal and vertical lines. Those lines format table
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are not
made or distributed for profit or commercial advantage and that copies bear
this notice and the full citation on the first page. Copyrights for components
of this work owned by others than ACM must be honored. Abstracting with
credit is permitted. To copy otherwise, or republish, to post on servers or to
redistribute to lists, requires prior specific permission and/or a fee. Request
permissions from Permissions@acm.org.
MOCR ’13 August 24 2013, Washington, DC, USA
Copyright 2013 ACM 978-1-4503-2114-3/13/08 ...$15.00.
contents to rows and columns. It is possible to estimate the
table structure directly from the image, but in this pap er it
is assumed that the table structure, which includes the ta-
ble header, the number of rows and columns, and even the
approximate cell size, is provided. The task of the table reg-
istration is matching the table model back to the scanned
image and then put ting the OCR results (words or para-
graphs) back to the table structure, such as a XML file, for
a better information organization. Figure 1 is an example
table image (a receipt) scanned at 300dpi resolution.
Figure 1: A typical receipt in China was s canned to
a 300dpi binary image.
1.1 Table Models
Based on the assumption that the table can be divided
to R rows and C columns, a table can be basically outlined
using R + 1 row lines and C + 1 column lines and all table
cells are rectangular. If a cell spans more t han 1 row or
column, then the table line segment inside of this cell will be
“erased”. Table models were provided as HTML (or XML)
files showing a series of cells with their specifications: row
index number, column index number, the number of rows it
spans, the number of columns it spans, cell height and cell
width. Some other extra information may also be provided,
such as if that cell is open on the left or right side. Based
on those table model files, tables can be fully reproduced.
For example, the given table mo del that matches the sample
image provided in Fig. 1 is plotted in Fig. 2.
1.2 Challenges
Even the table model is provided, there are several factors
that make the table registration a challenging task.
1.2.1 Cell Size Variance