Contributed Talk


Multiresolution orientation estimation using Riesz based operators


Martin Reinhardt
TU Bergakademie Freiberg, Germany

Abstract

Local orientation estimation is an important task in several elds of image processing like pattern recognition and texture analysis. Classical operators as the structure tensor or the Teager Kaiser energy operator rely on the quality of differentiation results which are usually noise corrupted or cause expensive computations. Therefore a pre-filtering approach is indispensable which leads to loss of information. The Riesz transform impacted signal processing applications during the recent years. It provides tools for image analysis in a multiresolution setting. Monogenic wavelet frames provide such techniques and allow fast computations in the Fourier domain. Our aim is to combine these two approaches to improve commonly used orientation estimators to give better results on real world data and also to reduce the loss of infor- mation.



ISSN 1611 - 4086 | © IKM 2015