Adaptive Multiscale Methods for Sparse Image Representation and Dictionary Learning

by Renato Budinich
Doctoral thesis
Date of Examination:2018-11-23
Date of issue:2019-01-16
Advisor:Prof. Dr. Gerlind Plonka-Hoch
Referee:Prof. Dr. Gerlind Plonka-Hoch
Referee:Prof. Dr. Armin Iske
Persistent Address: http://dx.doi.org/10.53846/goediss-7221

 

 

Files in this item

Name:tesi-render.7.12.18-fastweboptimized.pdf
Size:9.88Mb
Format:PDF
Description:Doctoral Thesis

The following license files are associated with this item:

Abstract

English

In this thesis we are interested in the topic of sparse digital image representation through adaptive multiscale basis. We develop for this two numerical methods: the Region Based Easy Path Wavelet Transform and the Haardict. The first method finds paths in regions of a segmented image and applies to these a wavelet transform. The second method uses a clustering procedure and the associated binary tree to define atoms for the dictionary learning problem which have the same structure as the wavelet coefficients of the classical Haar transform for one-dimensional signals. Both methods are analyzed in detail and numerical experiments proving their viability are discussed.
Keywords: multiscale adaptive basis; clustering; wavelet transform; image segmentation; region of interest; dictionary learning
 
Statistik