Illustration of the combination of image components, comparison with computational models of component combinations, and receptive field comparison.

<p><b>A</b> Image patch (bottom left) showing an intersection of two branches extracted from a grey-level natural scene image (adapted from the van Hateren natural image database <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1003062#pcbi.1003062-vanHateren1" target="_blank">[58]</a> with permission from J. H. van Hateren). Preprocessed version of the image patch (bottom right) obtained by using a center-surround filter to model the preprocessing by the lateral geniculate nucleus. <b>B</b> Left: Two image patches manually generated from the grey-level patch in <b>A</b>. Each patch is dominated by one of the two crossing branches of the original patch. Middle: The preprocessed versions of the two patches (central parts). Right: Combination of the two preprocessed patches using an occlusive combination (top) and a standard linear combination (bottom). <b>C</b> Examples of globular and Gabor-like receptive fields measured in V1 of macaque monkeys (courtesy of D. Ringach), and examples of the two receptive field types predicted by the occlusive encoding model. <b>D</b> Percentages of globular receptive fields predicted by different models for hidden units compared to percentages of globular fields of <i>in vivo</i> recordings.