Segmentation of heterogeneous document images : an ... - Tel
Segmentation of heterogeneous document images : an ... - Tel
Segmentation of heterogeneous document images : an ... - Tel
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
tel-00912566, version 1 - 2 Dec 2013<br />
Figure 2.12: Left image shows segmented lines from a <strong>document</strong> in [7] <strong>an</strong>d Right<br />
image displays projection pr<strong>of</strong>iles <strong>of</strong> vertical strips.<br />
segmentation contests.<br />
Hough-based methods c<strong>an</strong> also be considered as projection based methods.<br />
They are based on Hough tr<strong>an</strong>sform but instead <strong>of</strong> projecting pixel values into<br />
x or y axis, they project intensities onto parameter’s space. In general, the<br />
purpose <strong>of</strong> Hough tr<strong>an</strong>sform is to find imperfect inst<strong>an</strong>ces <strong>of</strong> objects within a<br />
certain class <strong>of</strong> shapes by a voting procedure. The classical Hough tr<strong>an</strong>sform is<br />
concerned with the identification <strong>of</strong> lines in <strong>an</strong> image, <strong>an</strong>d text lines in the case<br />
<strong>of</strong> <strong>document</strong> image. However, it is not limited to lines, <strong>an</strong>d <strong>an</strong>y kind <strong>of</strong> shape<br />
c<strong>an</strong> be found inside <strong>an</strong> image. In <strong>document</strong> image <strong>an</strong>alysis, Hough tr<strong>an</strong>sform<br />
is used in a variety <strong>of</strong> situations. Some methods use it to detect text lines or<br />
the skew <strong>an</strong>gle <strong>of</strong> the text lines, or to drive different characteristics such as the<br />
direction <strong>of</strong> a connected h<strong>an</strong>dwritten word, or even to detect table lines. Here<br />
we note two methods that use Hough tr<strong>an</strong>sform specifically for detecting text<br />
lines.<br />
In [58], G. Louloudis et al. have proposed a Hough tr<strong>an</strong>sform based h<strong>an</strong>dwritten<br />
segmentation method. Their earliest results have been published in<br />
[56, 57]. The method is adapted to deal with challenges <strong>of</strong> h<strong>an</strong>dwritten <strong>document</strong>s,<br />
including arbitrary sl<strong>an</strong>ted <strong>an</strong>d skewed text lines, accent marks above<br />
or below the text lines <strong>an</strong>d touching lines. After dividing the <strong>document</strong> image<br />
into equally sized blocks, the aim <strong>of</strong> Hough tr<strong>an</strong>sform becomes to find domin<strong>an</strong>t<br />
direction <strong>of</strong> connected components in each block. The algorithm also keeps<br />
track <strong>of</strong> the domin<strong>an</strong>t direction <strong>of</strong> all components inside the <strong>document</strong>. To form<br />
text lines, for each connected component a decision is held based on rules that<br />
compare the direction <strong>of</strong> each component in a block with its adjacent blocks.<br />
An additional constraint is applied upon which, a text line is valid only if the<br />
corresponding skew <strong>an</strong>gle <strong>of</strong> the line deviates from the domin<strong>an</strong>t direction by<br />
30