Projection and Hough-based methods are suitable for clearly separated lines. Projection-based
methods can cope with few overlapping or touching components, as long text lines smooth
both noise and overlapping effects. Even in more critical cases, classifying the set of blocks
into “one line width” blocks and “several lines width” blocks allows the segmentation process
to get statistical measures so as to segment moresurely the “several lineswidth” blocks. As a
result, the linear separator path may cross overlapping components. However, more accurate
segmentation of the overlapping components can be performed after getting the global or
piecewise straight separator, by looking closely at the so crossed strokes. The stochastic
method naturally avoids crossingoverlapping components (if they are not too close): the
resulting non linear paths turn around obstacles. When lines are very close, grouping methods
encounter a lot of conflicting configurations. A wrong decision in an early stage of the
grouping results in errors or incomplete alignments. In case of touching components, making
an accurate segmentation requires additional knowledge (compiled in a dictionary of possible
configurations or represented by logical or fuzzy rules).