Browse Prior Art Database

Character Segmentation Method for Japanese Documents

IP.com Disclosure Number: IPCOM000118269D
Original Publication Date: 1996-Nov-01
Included in the Prior Art Database: 2005-Apr-01
Document File: 2 page(s) / 56K

Publishing Venue

IBM

Related People

Yamashita, A: AUTHOR

Abstract

This method can segment Japanese characters including mixture size fonts (full, half, double, superscript, and subscript size fonts) from document images. The method first segments a character line image into small primitives represented as rectangles surrounding black pixels, and then detects an optimal combination of those primitives by searching the graph of primitives with an appropriate cost function.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 82% of the total text.

Character Segmentation Method for Japanese Documents

      This method can segment Japanese characters including mixture
size fonts (full, half, double, superscript, and subscript size
fonts) from document images.  The method first segments a character
line image  into small primitives represented as rectangles
surrounding black pixels,  and then detects an optimal combination of
those primitives by searching  the graph of primitives with an
appropriate cost function.

      The Figures show an example applied by the method.  Fig. 1
shows extracted primitives (1).  Base lines, standard character size,
and column pitch are estimated from those primitives in a character
line (2), then appropriate labels such as a part of full size font,
connected half size font, and superscript font are assigned to each
primitive (3).  Each label associates size and location cost
calculated from the relative position and size to the base line and
the standard font size (4).

      Fig. 2 shows possible combinations of primitives represented as
a directed graph.  A node of the graph represents a tentative
character which consists of one or more primitives (1).  The cost
function is defined as summation of different costs; the size and
location cost, the  pitch cost calculated from the linked three
nodes, and recognition cost  obtained by recognizing a node as a
character image.  The path with the  minimal cost is selected as an
optimal segmentation result (2).

    ...