In the next step, the mean shift segmentation [19] is carried
out on P to identify the candidate graphical regions of the
document image. Let the segmented image be S, and r be
the total number of segments present in S. Smaller segments
are ignored at this stage, and a threshold value on the sizes
of segments is used for this purpose. The largest segment is
considered to be background B. The rest of the segmented
regions form the set of candidates for graphical regions. Let
this set be denoted by F: