Figure - PMC

Skip to main content

View full-text article in PMC

. Author manuscript; available in PMC: 2013 Mar 18.

Published in final edited form as: IEEE Trans Image Process. 2011 Jun 20;21(1):130–144. doi: 10.1109/TIP.2011.2160072

Fig. 7 — (Left) CS results on gray-scale Castle image, based on DCT dictionary D. CS projection matrix Σ is randomly constituted, with elements drawn i.i.d. from (0,1). Results are shown using the DP-BPFA and PSBP-BPFA models in Section IV. Comparisons are also made with several CS inversion algorithms from the literature. (Right) Same as on the left but based on learned dictionary D, instead of DCT. The online BP results employ BPFA to learn D and do CS inversion jointly. All other results are based upon learned D with learning performed offline using distinct training images.

Inline graphic — (Left) CS results on gray-scale Castle image, based on DCT dictionary D. CS projection matrix Σ is randomly constituted, with elements drawn i.i.d. from (0,1). Results are shown using the DP-BPFA and PSBP-BPFA models in Section IV. Comparisons are also made with several CS inversion algorithms from the literature. (Right) Same as on the left but based on learned dictionary D, instead of DCT. The online BP results employ BPFA to learn D and do CS inversion jointly. All other results are based upon learned D with learning performed offline using distinct training images.