Fig. 4.

The inferred dictionary for the synthesized data in Figure 3(b). (a) dictionary elements at layer one, ${\mathit{d}}_k^{\left(1\right)}$ and layer two, ${\mathit{d}}_k^{\left(2\right)}$; (b) estimated posterior distribution on the number of needed dictionary elements at level two, based upon the Gibbs collection samples. In (a) the images are viewed in the image plane. The layer one dictionary images are 4 × 4, and the layer one dictionary images are 9 × 9 (in the image plane, as shown). In all cases white represents one and black represents zero.