Browse Prior Art Database

Measurement of Features in Text Images

IP.com Disclosure Number: IPCOM000102701D
Original Publication Date: 1990-Dec-01
Included in the Prior Art Database: 2005-Mar-17
Document File: 2 page(s) / 95K

Publishing Venue

IBM

Related People

Wendt, P: AUTHOR

Abstract

Disclosed is a means to measure the sizes and thicknesses of features in binary and gray-level images by using morphological operators and related nonlinear filters.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Measurement of Features in Text Images

       Disclosed is a means to measure the sizes and thicknesses
of features in binary and gray-level images by using morphological
operators and related nonlinear filters.

      In a binary image (e.g., text), the regions in the foreground
have value 1, and the background has value 0.  A morphological
opening (1), with a structuring element that is a circle of radius R,
can filter out detail in the foreground of dimension < 2R.  A set of
n such openings, with radii R1 < R2 < ... Rn, and operating in
parallel, can distinguish a range of levels of detail.

      Addition of the outputs of the n openings produces a gray-level
"thickness" image.  For a pixel in the background of the original
image, the value in the sum image is 0. Otherwise, it increases with
the local thickness of the stroke or feature to which the pixel
belongs.  In fact, for a given pixel in a region in the foreground,
it is the index of the largest structuring element that can contain
the pixel and fit in the region at the same time.

      In fact, the sequence of structuring elements can be arbitrary,
so long as it is increasing (i.e., the i'th structuring element is
contained in the (i+1)'st, for all i).  This allows the size
measurement to be directional, or sensitive to shape.

      For a gray-level image, we can threshold it, then process it as
above.  The thresholds for the different openings can be different,
so long as they do not decrease with increasing size of the
structuring elements.

      For more robustness to noise and image imperfections, we can
replace the openings with more general filters.  If the maximum
possible pixel value corresponds to black (i.e., 1 or 255), an
opening consists of a local minimum on a window defined by a
structuring element, followed by a local maximum on the same window.
We can replace this min/ max pair by another dual pair of
ranked-order filters - i.e., a k'th- smallest/k'th-largest filter
pair, for some k.  Then, the first filter will tolerate up to k-1
bright pixels in a window before its output becomes bright.  As the
two filters have symmetric "ranks", the shrinking and expansion
produced by the filters are relatively symmetric for details that
survive the shrinking.

      Such filters make the whole measuring scheme resistant to
bright pixels in dark regions, that arise from faded documents or
from bad print quality,...