Human visual system easily and rapidly recognizes a scene or image under different affine transformations, which is not the true for the machine. Rotation is more complex than translation and engenders more difficulties in analysis. This paper address evaluation and comparison of texture descriptors, particularly Local Relational String, under rotation effects. Many methods are invariant for geometric transformation, but this is not sufficient to handle the classification problem. We show in this study, when training samples represent a large range of rotated textures, methods with high discriminative properties leads to a very good classification rate despite their no invariance for rotation.
展开▼