bin_tfj

Name

bin_tfj - binarize a gray-scale band using Taxt, Flynn and Jain`s method

Syntax

 #include <xite/binarize.h>

 int bin_tfj( IBAND inband, IBAND outband,
    int region_size, int step_size,
    int training_size );

Description

The input band inband is binarized using the method of Taxt, Flynn and Jain. The result is returned in outband.

Taxt, Flynn and Jain`s method divides the image into non-overlapping windows of size (small_window_size x small_window_size). For each window, a larger concentric window of size (region_size x region_size) pixels. The histogram of each of the larger windows is approximated by a mixture of two Gaussian distributions. The parameters of the mixture are estimated using an expectation-maximization (EM) algorithm. In each small window, the pixels are classified using the quadratic Bayes classifier.

A training set of training_size pixels are picked on random, but edge pixels are avoided in the training set. The EM algorithm requires global start values for the estimated class means mu1 and mu2, and estimated standard deviations sigma1 and sigma2. These are obtained by k-means clustering of the whole input band. The EM algorithm also needs an initial estimate of the mixing weight pi, and pi = 0.5 is used.

Restrictions

inband and outband must have pixel type unsigned byte.

Return value

 0 : ok
 1 : Bad input pixel type
 2 : Bad output pixel type

References

T. Taxt, P.J. Flynn & A.K. Jain: "Segmentation of document images", IEEE Transactions on Pattern Analysis and Machine Intelligence, vol T-PAMI 11, no 12, pp 1322-1329, 1989.

Files

src/binarize/bin_tfj.c

Author

Řivind Due Trier

Id

$Id: bin_tfj.c,v 1.15 1997/01/14 15:46:57 svein Exp $