Highly Nonlinear Approximations for Sparse Signal Representation


Analysis of the Self Projected Matching Pursuit Algorithm

The paper below presents the convergence and numerical analysis of the Self Projected Matching Pursuit (SPMP) approach, which is a low memory implementation of the Orthogonal Matching Pursuit method. The approach provides an iterative way of solving the least squares problem with much less storage requirement than direct linear algebra techniques. Hence, it is appropriate for solving large linear systems. The Hierarchized Block Wise (HBW) version of SPMP (HBW-SPMP), for approximating a signal partition subjected to a global constraint on sparsity, is considered and illustrated by producing high quality sparse approximation of nonstationary music clips.

"Analysis of the Self Projected Matching Pursuit Algorithm"
by Laura Rebollo-Neira, Miroslav Rozloznik, and Pradip Sasmal

The MATLAB routines for implementing the methods are available here.