< Back to Article

Cortical Surround Interactions and Perceptual Salience via Natural Scene Statistics

Figure 4

Visualization of the covariance matrices learned from scenes between the Gaussian variables associated with center and surround RF outputs in the mixture components of Fig. 3.

Top row: co-assigned components (from left to right, ); bottom row: independent component (). Black bars denote the orientation and relative position of the RFs; bar thickness is proportional to the variance. The thickness of the red lines connecting pairs of bars is proportional to the absolute value of the covariance. For each surround group, we show only the covariance with the center of the same orientation; the covariance with center RFs of different orientations is one to three orders of magnitude weaker. Similarly, we show only the covariances between RFs with even phase; the covariances for the odd phase are similar, while those across different phases are negligible. In each mixture component the variances of the center RF and its collinear neighbors, as well as the covariance between them, are larger reflecting the predominance of collinear structure in scenes.

doi: https://doi.org/10.1371/journal.pcbi.1002405.g004