M.K. Bashar, Y. Takeuchi, T. Matsumoto, and N. Ohnishi (Japan)
: Cortex transform, texture, mean energy, directional surface density, integration, confusion matrix.
We propose a scheme of integrating two feature groups
computed with and without transformation of the original images. Transform features are obtained by popular
cortex transform technique, while contrast features are
computed from the original intensity images. we pro
pose three contrast features, namely, directional surface
density (DSD), normalised sharpness index (NSI), and
normalized frequency index (NFI) as measures for pixel
brightness variations. Fusion by simply stacking vectors
as well as by correlation is performed in the feature space
and then classiﬁcation is done using minimum distance
classiﬁer on the fused vectors. Transform features extract
smoother texture boundaries in micro-textured images
(natural scenes) while contrast features better extract
boundaries in highly (regular) textured images(mosaic
images). This inverse properties are combined through
vector fusion for robust texture classiﬁcation of two image groups namely mosaic and natural scenes obtained
from Brodatz album and VisTex database respectively.
Error matrix and edge-smoothness analysis show the ro
bustness of the proposed scheme.