A multiscale transform in medical imaging
My name is Dominic Amann and I am working in the transfer group at the Johann Radon Institute for Computational and Applied Mathematics in Linz. My research topic is image processing for medical purposes such as optical coherence tomography.
The goal of our work is to remove the noise that is inherent to many medical imaging techniques while retaining and even enhancing the underlying structures. Finding a balance between these two conflicting demands depends on many factors such as the image source, intended purpose of the processed image and user preferences. Therefore a second goal is to identify a set of intuitive parameters that can be exposed to the users so that they may modify the algorithm on the fly to fit their needs.
The core of our algorithm is a multiscale transform with needle-shaped elements of many directions and positions at each scale. This allows for good representation of objects with smooth regions separated by edges.
This means the desired structural information is contained in only a few coefficients while the noise is distributed among many which can be used for denoising. Furthermore, the multiscale representation facilitates post processing by allowing us to analyse and modify coarser and finer image information separately.
Reblogged this on MI-NET.