Figure 1. Sequence of images in an experimental trial.

Figure 2. Noise contrasts for the experiments (edge present). Noise contrast of 0.04 Michelson contrast units (left), (b) 0.148 (middle).

Figure 3. Counter to our predictions, summation fields were found to expand by a factor of 2-3, primarily in the X direction, across the edge stimulus.

Figure 4. Maximum Likelihood Gaussian model fit to the classification image data. The width of the X-profile increases as a function of noise contrast.

Edge Detection and Classification Images

(For complete details, see our VSS 04 poster).

Detection of low-contrast luminance-defined stimuli can involve spatial summation over a large portion of the visual field. However prior psychophysical results suggest that the summation region may shrink substantially in the presence of high-contrast masking gratings or noise (Legge & Foley, 1980; Kersten, 1984). This may be related to recent findings that the extent of spatial summation in V1 neurons depends upon contrast (Sceniak et al., 1999; Kapadia et al., 1999). Here we use a classification image technique to directly test whether the psychophysical receptive field for a simple stimulus (a vertical edge in noise) is dependent upon contrast. Classification images for yes/no edge detection were estimated at noise contrasts ranging from 4%-50%.

See Figure 1.

Trial sequence:

1. Fixation
2. Blank Screen
3. Edge + Noise OR Noise alone

Observer indicated presence or absence of the edge by pressing 1 of 2 keys.

See Figure 2.

Results:
Counter to our predictions, summation fields were found to expand by a factor of 2-3, primarily in the X direction, across the edge stimulus.

See Figure 3.

Edge detection summation fields were found to be well-approximated by a first-derivative of Gaussian model. X-Y separability of the model was used to compute 1D X and Y projections of the classification image data (Figure 4). This revealed an expanding X-profile as noise contrast was increased.

See Figure 4.

Summary of Results:

1. Edge detection summation fields are well-approximated by a first derivative of Gaussian model.
2. Summation fields were found to expand by a factor of 2 to 3 with increasing noise contrast. This expansion appears to be primarily across the edge, not along the edge.
3. These results run counter to prior physiological and psychophysical data suggesting a contraction of receptive field size with increasing contrast.
4. Summation fields were generally found to be displaced toward the lower hemifield, particularly at higher contrasts.