Figure-Ground Organization in Visual Cortex for Natural Scenes

Responses from two example cells to square, full natural scene and patch of natural scene. Four frames are shown for each stimulus type, with the two sides of border-ownership shown side by side and the two contrast polarities at top and bottom. Red circles indicate location and approximate size of the CRF. The patch stimuli have been magnified for illustration. Red frames indicate stimuli with objects on the preferred side, and blue frames indicate stimuli on the opposite side. Only one of the presented scenes is shown in each case. Cell 1 was presented 44 scenes, and cell 2 was presented 177 scenes. The mean temporal responses after the onset of the stimulus are plotted below with the corresponding colors; shading indicates 95% confidence intervals.

Comparison of border-ownership signals produced by natural scenes and squares. For each of the 140 V2 cells studied, the border-ownership effects of full image and patch, and the context influence, are plotted against the border-ownership effect of squares. The effects were measured by linear regression performed on square root–transformed spike counts (see Materials and methods). Error bars bracket the range of 6× the mean standard errors of estimates. The slopes of the lines, determined by minimizing the orthogonal squared deviations, indicate the relative strengths of border-ownership signals for the natural scene stimuli relative to squares across the population. Shaded areas and dashed lines indicate 95% confidence intervals. Colors mark the example neurons of Fig. 2 (red, Cell 1; blue, Cell 2).

Variation of border-ownership signals across scenes. Data from the same cells as in Fig. 2. Scenes were sorted, in decreasing order, by strength of border-ownership signal. Black lines represent signals for full image, patch, and context. Gray bands show the 95% confidence intervals of the sorted effects obtained under the null hypothesis (no border-ownership selectivity). Horizontal blue line indicates the border-ownership signal for squares. The 10 scenes that elicited the most positive and the 10 scenes that elicited the most negative border-ownership signals are shown for each cell and condition in Figs. 4.1 and 4.2.

Consistency and latency of neurons in signaling border-ownership for natural scenes. A, “Raw” shows the distribution of the proportion of scenes for which a cell gave the same sign of border-ownership signals. “Corrected” shows the distribution of the proportion after correction for random variation (see Results). Cells are selected for significant (p < 0.01) effect of border-ownership in full natural scenes (n = 65). B, Estimates of the latencies of the border-ownership signals for natural scenes in the individual cells plotted as a function of their consistency. The latency of the border-ownership signals did not increase with consistency.

Time course of border-ownership signals for squares and objects in natural scenes. A, The curves show smoothed poststimulus time histograms of the mean border-ownership signals, averaged over cells, for full natural scene (red), natural scene patch (blue), natural scene context (green), and square (black dashed). The border-ownership signal for natural scenes and its context component emerge at the same time as the signal for square figures, but rise slightly more slowly. The patches of natural scenes produce only a small transient signal. B, Comparison of the time course of border-ownership signal (red) and mean response (black) for contours of natural scenes. Note the short interval between response onset and rise of the border-ownership signal.

Determination of the latencies of the population border-ownership signals. Red curves show the cumulative spike difference counts for small and large squares, the full natural scenes, and the context component. The latencies were determined by two-phase regression fits (blue lines; see Materials and methods).