The segmentation of a document image in WISDOM++

The segmentation of the document image into rectangular blocks enclosing content portions is performed by means of a variant of the Run Length Smoothing Algorithm (RLSA).

The original algorithm (Wong et al., 1982) applies four operators to the document image:

  • Horizontal smoothing with a threshold Ch
  • Vertical smoothing with a threshold Cv
  • Logical AND of the two smoothed images
  • Additional horizontal smoothing with another threshold Ca.
  • and, although conceptually simple, requires scanning the image four times.

    WISDOM++ implements a variant where:

  • the image is scanned only twice with no additional cost (Shih& Chen, 1996);
  • the smoothing parameters Cv and Ca are adaptively defined on the ground of the spread factor;
  • the segmentation is sped up since it is performed on a document image with a resolution of 75 dpi (reduced document image).




    Back to home page