Image Processing in Scale-Orientation Space

Scale-orientation signature space represents a non-linear transformation from normal grey-level image space. From this point of view there are similarities with other representations such as radial basis functions, Fourier, Mellin and wavelet transforms. Morphological related techniques are used to generate the scale-orientation space. A series of filtering techniques, based on selectively removing values in scale-orientation space, have been developed. Transportation across scales, but not orientations, in scale-orientation space is discussed as a way to preserve the base of image structures. Evaluation of the described techniques is based on a set of standard images which contain linear structures superimposed on a complex texture background. Both quantitative and qualitative results are presented.