Segmentation of heterogeneous document images : an ... - Tel

(a) Five zones 1-5 (b) Projection profile
of zone 3
(c) first derivative (d) initial, refined
and final regions
Figure 2.13: Steps for locating text line separators in part of document image. [75]
tel-00912566, version 1 - 2 Dec 2013
less than 2 ◦ . This method is successful provided that the free parameters are
set correctly. Furthermore, because the algorithm keeps track of the dominant
direction, the whole document must have text lines with roughly the same direction.
The method published in [61] is one more method based on Hough transform.
In the first step, authors apply a Hough transform to each connected
component namely the handwritten words to find the direction of a component.
Then the algorithm searches for the nearest neighbors of each component in four
principal directions. Once the neighbors are found, a weighted directed graph
is built by connecting each component to its neighbors with a weighted edge
proportional to the geometric distance between components. Finally, to form
text lines, the algorithm removes top to bottom edges based on thresholding
the length of edges.
Texture based methods
Any method that is based on some kind of filtering, shall it be Gabor, Wavelet,
Gaussian or just the averaging operator can fit into this category.
The first method that we review is for text line segmentation from freestyle
script-independent handwritten or printed documents. Y. Li et al. first have
published their preliminary results for this method in [52] and later in [53]. For
this method it is assumed that text lines have a horizontally elongated shape,
but still a variation of ±10 ◦ is allowed. The method estimates a probability density
function based by convolving the image with a non-parametric anisotropic
Gaussian kernel. The initial estimates of the text line boundaries are computed
by thresholding this density function map and then a level set method evolves
from the initial estimations to obtain the final text line boundaries.
Another method is proposed in [29]. Du et al. propose a script-independent
method for segmentation of handwritten text lines based on a piecewise ap-
31