Figure 9.

Comparison with previous analyses of optical SF responses. A, An interpolated SF map constructed using the techniques described in Materials and Methods. B, An interpolated SF map constructed by first averaging over all orientations and then interpolating between SFs to get the preferred SF. SF pinwheels similar to those found by Everson et al. (1998) can be seen in this map. C, An SF response map constructed by selecting the preferred SF at the preferred orientation. Pixel lightness is proportional to response strength. This map uses blank-screen normalization with no averaging over orientation. D, An SF response map constructed by normalizing single-condition images to a cocktail blank, averaging the response of a pixel over all orientations and then selecting the SF that most strongly activated the pixel [similar to the protocols of Hubener et al. (1997) and Shoham et al. (1997)]. Pixel lightness is proportional to response strength. There are fewer bright pixels at intermediate SFs than there are in the map of C. E, Response amplitude as a function of preferred SF for the map shown in C. The red line with open circles shows the distribution of average response amplitudes as a function of spatial frequency. Note that responses are greatest for intermediate SFs. The blue line with filled symbols is calculated from a subset of the data containing all but the highest two SFs (1.2 and 1.84 c/°). Note that response intensities for the truncated and complete data set are nearly identical. F, Response amplitude as a function of preferred SF for the map shown in D. The dashed red line with open triangles shows the distribution of average response amplitudes as a function of spatial frequency. Response amplitudes are weakest for intermediate SFs, unlike the profile in E. The depth of the trough shown in the profile depends on the noise threshold used to generate the SF map; maps generated using a lower threshold, and therefore excluding fewer points from analysis, would have a response profile with a shallower trough. The dashed blue line is calculated from a subset of the data containing all but the highest two SFs (1.2 and 1.84 c/°; using the same threshold level) and shows a peak at 0.8 c/°. This profile is completely different from the one shown in red, calculated from the entire data set. Cocktail-blank normalization therefore makes the response profile dependent on the particular stimulus set used for analysis.