![](bg6.png)
We tested our algorithm on a mix of several
educational and business forms containing different
types of ICR cells used in different layouts. Our
batch consisted of 918 different images, which were
further divided into two separate sets based on
whether template information is present. We used
Newgen OmniExtract Form Processing Engine to
run our tests. Caere engine was used for ICR.
On a batch size of 500 images, structured form
processing approach was followed that used
template information. We tested using both the
approaches; the traditional vector distance mapping
and our proposed approach. The image dataset had a
collection of images with skew (+ 3 degrees), shift
and shrinkage. Out of the 500 images, 10% of the set
had images that contained broken or missing cells
and required estimation. We recorded a 77%
improvement in data extraction using the proposed
algorithm. We calculated the number of correctly
extracted ICR cells for both the approaches to get a
measure of the improvement in data extraction. The
improvement was due to the accurate ICR cell
detection and estimation and better form removal
(Fig. 8).
On a batch size of 418 images, we followed the
unstructured form processing approach by passing
the whole image to our engine as input without any
template information. The results were again very
promising with a total data extraction accuracy
percentage of 97.9%.
The experiment results show that our approach,
even when not using the template information,
brings in highly accurate data extraction results
when compared to the traditional form processing
approach. The result also underscores the fact that
the proposed solution can be applied to unstructured
form processing where ICR cells can be detected
and used for document understanding, classification,
and segmentation.
REFERENCES
Liu J., Ding X., Wu Y. 1995. Description and recognition
of form and automated form data entry. In ICDAR'95,
Third International Conference on Document Analysis
and Recognition - Volume 2,pp. 579-582.
Mathur, A., Gur, N.H., 1999. High Performance Form
Analysis and Data Extraction. In ICDAR’99, Fifth
International Conference on Document Analysis and
Recognition.
Pitas,I. and Venetsanopoulos, A. N. , 1990. Nonlinear
DigitalFilters. Boston: Kluwer Academic, 1990.
Chih-Hong, K., Hon-Son, D., 2005. Skew Detection of
Document Images Using Line Structural Information,
In ICITA'05, Third International Conference on
Information Technology and Applications Volume 1.
Shi, Z., Govindaraju, V., 2003. Skew Detection for
Complex Document Images Using Fuzzy Runlength.
In ICDAR'03, Seventh International Conference on
Document Analysis and Recognition - Volume 2, 715-
719.
Le D. X., Thoma G.R., Wechsler H.1996. Automated
border detection and adaptive segmentation for binary
document images. In Proceedings of ICPR ’96.
Illingworth, J., Kittler J., H.1998. A survey of the Hough
transform. CVGIP, Vol. 44.87-116, 1998.
Rosito Jung C., Schramm R. 2004. Rectangle detection
based on a windowed Hough transform. In
SIBGRAPI'04, Computer Graphics and Image
Processing, XVII Brazilian Symposium on , 113-120.
Zheng Y., Li H., Doermann D. 2003. Background line
detection with a stochastic model. In 2003 Conference
on Computer Vision and Pattern Recognition
Workshop - Volume 3.
Gattani, A., Mukerji, M. and Gur, H., 2003. A Fast
Multifunctional Approach for Document Image
Analysis. In Proceedings of the Seventh ICDAR, 2003,
1178-1182.
Yoo J., Kim M., Yong Han S. 1995. Line Removal and
restoration of Handwritten Characters on the Form
Documents. In ICDAR'97, Fourth International
Conference Document Analysis and Recognition ,
128-131.
Dillencourt, M.B., Samet,H., and Tammininen,M., 1992.
General approach to Connected-Component Labeling
for Arbitrary Image Representations, In J.ACM Vol
39, No.2, 1992, pp. 253-280.
VISAPP 2006 - IMAGE ANALYSIS
276