Problem
Consider a set of signals fY1; Y2 : : : ; YLg ⊂ IRN, Suppose that there exits an overcomplete dictionary D 2 IRN×M with M ≥ N such that each signal can be sparsely
approximated under the dictionary D. The dictionary learning aims at find such a
dictionary so that each signal Y‘ can be well approximated by Dc‘ where c‘ 2 IRM
is a sparse vector. The dictionary learning problem can be formulated as solving the
following optimization problem
min
D2IRN×M;C2IRM×L
kY – DCk2 F; (1)
subject to
kDjk1 = 1; for j = 1; 2; : : : ; M;
kC‘k0 ≤ K; for ‘ = 1; 2; : : : ; L:
As the algorithm cannot guarantee the convergence. The algorithm usually stops
after R (as an input) iterations.
Goal
In Lecture notes, The K-SVD method is presented for solving dictionary learning
problem. The goal of this project is to implement the K-SVD method with possible
modifications to have your version of the dictionary learning method that solves (1