10.03.2015 Views

sparse image representation via combined transforms - Convex ...

sparse image representation via combined transforms - Convex ...

sparse image representation via combined transforms - Convex ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

156 APPENDIX B. FAST EDGELET-LIKE TRANSFORM<br />

1 2 3<br />

4 X-interpolate<br />

5 =⇒<br />

6<br />

1 2 3 4 5 6<br />

Figure B.1: X-interpolation.<br />

For 0 ≤ k, l ≤ N − 1, we have<br />

=<br />

=<br />

=<br />

( k<br />

F<br />

N − 1 2 2)<br />

,s(k)+l · ∆(k) − 1<br />

∑<br />

i=0,1,... ,N−1;<br />

j=0,1,... ,N−1.<br />

∑<br />

j=0,1,... ,N−1<br />

∑<br />

j=0,1,... ,N−1<br />

I(i +1,j+1)e −2π√ −1( k N − 1 2 )i e −2π√ −1(s(k)+l·∆(k)− 1 2 )j<br />

⎛<br />

⎝<br />

= e −2π√ −1∆(k) l2 2 ·<br />

∑<br />

i=0,1,... ,N−1<br />

⎞<br />

I(i +1,j+1)e −2π√ −1( k N − 1 2 )i ⎠<br />

} {{ }<br />

define as g(k, j)<br />

g(k, j)e −2π√ −1(s(k)+l·∆(k)− 1 2 )j<br />

∑<br />

j=0,1,... ,N−1<br />

e −2π√ −1(s(k)+l·∆(k)− 1 2 )j<br />

g(k, j)e −2π√ −1(j·s(k)− 1 2 j+∆(k) j2 2 ) · e π√ −1∆(k)(l−j) 2<br />

.<br />

} {{ }<br />

convolution.<br />

The summation in the last expression is actually a convolution.<br />

The matrix (g(k, j)) k=0,1,... ,N−1 is the 1-D Fourier transform of the <strong>image</strong> matrix I(i +<br />

j=0,1,... ,N−1<br />

1,j+1) i=0,1,... ,N−1 by column. We can compute matrix (g(k, j)) k=0,1,... ,N−1 with O(N 2 log N)<br />

j=0,1,... ,N−1<br />

j=0,1,... ,N−1<br />

work. For fixed k, computing function value F ( k N<br />

,s(k)+l · ∆(k)) is basically a convolution.<br />

To compute the kth row in matrix ( F ( k N ,s(k)+l · ∆(k))) k=0,1,... ,N−1 , we utilize the Toeplitz<br />

l=0,1,... ,N−1<br />

structure. We can compute the kth row in O(N log N) time, and hence computing the first<br />

half of the matrix in (B.4) has O(N 2 log N) complexity.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!