Generalized Eigen Cooccurrence : Application to Palaeography
Résumé
This paper introduces the Generalized Eigen
Cooccurrence Matrix (GECM) as a new feature to des
cribe complex structures like images of handwritings for
palaeographic expertise. It measures the spatial de
pendency between two features in the image. It generalizes t
he popular grey level cooccurrence Dependencies (SGLD) which u
ses the luminance for the two features. 2nd order statistic
s generate high dimensional feature space which must be reduced to
overcome the curse of dimensionality. Haralick have described several
descriptors suited for SGLD matrices that cannot be used in
Generalized Cooccurrence. In our case, the cooccurrence matrices are not always symmetric and the contents of each matrice are different from the SGLD. We introduce t he GECM
which uses the eigen decomposition of the cooccurrence matrices
to reduce the number of matrices and decrease the redundancy
of spatial information instead to reduce the size of each matrix.
We show the effectiveness of the GECM on palaeography
application and writing comparison.