Rotation and scale invariant classification has become
a well established
area of texture analysis. However, content-based image retrieval often requires
a higher degree of invariance, since a pattern may appear
in a wide range of 3D orientations. This is a new challenge for the existing
approaches to texture. It seems that most of them are not prepared to
face it.
Recently, we have proposed a general measure of pattern regularity [
5]
that is stable under weak perspective, i.e., orthographic projection plus scaling.
In this paper we extend this measure to a feature vector and apply the
new approach to invariant classification of regular
textures under orthographic projection. Accuracy over
is achieved for
18 patterns.
