Morphological filtering via energy minimization

Emrah Bostan

Semester Master Project

Microengineering section, EPFL

June 2010

Abstract

In this study, knowing that the total variation based denoising methods result in sharp edge-preserved images and that L1+TV is a contrast invariant filter, a new algorithm for L1+TV denoising has been developed and implemented.
The algorithm, which is a Majorize Minimize (MM) approach, further takes the advantage of a proper usage of the matrix inversion lemma for the specific denoising purpose.
This new algorithm provided an iterative solution for the successive quadratic energy minimizations which cannot be performed by direct matrix inversion due to computational performance of present day.