With this operation, text regions of a document are detected,
and most of edge information from characters on a document
are removed in each edge plane. Fig.6 shows some results
of removing text regions from edge planes. The proposed
method of removing text regions preserves line segments in
each decomposed edge plane, while it removes text regions
effectively. Even though the result of the detected text regions
are not absolute perfect as shown in Fig.6, it eliminates a lot
of outliers from characters for the Hough transform in the next
step to extract line features of a paper boundary.
A homography could be estimated, if four corresponding
points on a plane between an image and the world are known.
To find these features, we try to find the boundary of a paper
using the Hough transform. Edge planes have already limited
range of edge angles. For example, the first decomposed edge
plane after removing text regions, 30, only has edge pixels
with the edge angles from 0 degree to 45 degree. The Hough
transform is applied in each decomposed edge plane after
removing text regions. It is the same that the Hough transform
is applied in a reduced angle domain. Because all of the
edge pixels on a boundary line of a paper has similar angle
values, it is more robust than applying a traditional the Hough
transform to a whole edge plane. One line segment in each
decomposed edge plane is selected as the result of the Hough
transform. Finally, four line features with higher rank among
the extracted eight line features from each decomposed edge
plane are found