Summary
International Symposium on Nonlinear Theory and its Applications
2008
Session Number:A3L-D
Session:
Number:A3L-D3
Sparse representation of L-order Markov signals
Zhaoshui HE, Andrzej Cichocki,
pp.-
Publication Date:2008/9/7
Online ISSN:2188-5079
Summary:
In engineering applications, many signals are non-white (i.e., they are colored and have temporal structure) and can be modeled as an L-order Markov process. However, the existing sparse representation methods do not consider the L-order Markov property of signals. To fill this gap, we propose a new sparse representation framework: Firstly, we segment (split) the available T samples into several frames, where the length of each frame is L(1 ? L ? T ); secondly, to make the estimated signals be smooth, we set an appropriate percentage of overlapping between two neighboring frames (typically, 50%- 70% overlapping); finally, we perform sparse representation for each frame. Under this framework, a modified-BP algorithm is developed by L-order lpq-norm-like optimization, which can indirectly exploit the L-order Markov property of sources and achieve better results.