Analysis of dictionary components learned by the SSMCA algorithm on natural image patches.

<p><b>A</b> Example dictionary elements <b>W</b>_<i>h</i> after learning. <b>B</b> Fraction of globular fields estimated from <i>in vivo</i> measurements, compared to ours (after fitting with Gabor wavelets and DoG’s; globular percentages taken from [<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0124088#pone.0124088.ref019" target="_blank">19</a>] who analyzed data provided by [<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0124088#pone.0124088.ref018" target="_blank">18</a>] and estimated percentages of globular fields from data in two further papers [<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0124088#pone.0124088.ref043" target="_blank">43</a>, <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0124088#pone.0124088.ref044" target="_blank">44</a>]. <b>C</b> Learned prior. <b>D</b> Actual activations of diverse dictionary elements <i>s</i>_<i>h</i> (posterior averaged over data points).