Structure Multivector for Local Analysis of Images

The structure multivector is a new approach for analyzing the local properties of a two-dimensional signal (e.g. images). It combines the classical concepts of the structure tensor and the analytic signal in a new way. This has been made possible by using a representation in the algebra of quaternions. The resulting method is linear and of low complexity. The filter-response includes local phase, local amplitude and local orientation of intrinsically one-dimensional neighborhoods in the signal. As for the structure tensor, the structure multivector field can be used to apply special filters to it for detecting two-dimensional features like corners. Experiments and comparisons with other approaches have been made.