The four kinds of 64 x 256 dictionaries being considered here include: (a) LARS + ASVD + ADDAPTIVEL denotes learning dictionaries by the proposed DL algorithm; (b) LARS + ASVD denotes learning dictionaries by algorithm, of which the sparse representation stage adopts LARS and the dictionary update stage adopts ASVD; (c) learning dictionaries by K-SVD algorithm; (d) a dictionary based on scaling and translation of pre-defined basis functions, RDCT.
From Figure 3, compared (b) with (c), the CS reconstruction performance with (b) is better than that with (c), while at most 10 coefficients are used for sparse representation stage in both of them, so we infer that the LARS in sparse representation stage and the ASVD in dictionary update stage is the reason for superiority of (b).
In the proposed DL algorithm, LARS with LASSO modification and adaptive sparsity constraint are used in the sparse representation stage, and ASVD is used in the following dictionary update stage.
An alternative technique for computing the ASVD is presented in .
A continuation algorithm for computing the ASVD is presented in .
The relevant ASVD, A(t) = U(t)[SIGMA](t)V[(t).sup.T], -0.5 [less than or equal to] t [less than or equal to] 0.5, can be computed explicitly:
The SupraExpress 336i Sp with ASVD (internal model) carries an estimated retail price of $129.
The SupraExpress 336 Sp with ASVD includes Blizzard's Warcraft II, a multiplayer strategy game for modem-to-modem gameplay.