The Journal of Neuroscience, July 25, 2007, ():
Benefits of Contrast Normalization Demonstrated in Neurons and Model Cells
J. Neurosci. Gaudry and Reinagel
27: 8071
Supplemental Data
Files in this Data Supplement:
- supplemental material
-
Supplementary Figure 1. Distribution of contrasts in natural scene stimuli. Across nine hours of time-varying natural stimuli from (van Hateren 1997), we calculated the distribution of contrasts and found that most stimulus epochs were characterized by contrasts between 11% and 100%. a. The probability of contrast across all natural-scene stimuli, where contrast was calculated for 500-ms windows. Contrast is defined as the standard deviation of the luminance divided by the mean; we note that contrast can be >100%. b. The probability across stimuli that the contrast, calculated for 500-ms windows, was less than 11% (red bar), between 11% and 33% (green bar), between 33% and 100% (blue bar), or greater than 100% (black bar). c. The probability that the contrast was less than 11% (red curve), between 11% and 33% (green curve), between 33% and 100% (blue curve), or greater than 100% (black curve) when contrast was calculated using a variety of window sizes.
- supplemental material
-
Supplementary Figure 2. Mutual information between visual stimuli and LGN responses. a. Mutual information I in units of bits/s for responses to 100% contrast stimuli (horizontal axis) compared to the same cell at either 33% contrast (○,�) or 11% (∆,▲) contrast. Filled symbols indicate cases in which the correction for finite data size was small and linear for both contrasts (see Methods). Symbols below the diagonal indicate that information decreased with contrast. Across the population the decrease from high to medium contrast and from high to low contrast was highly significant (p<0.01 for both 100% to 33% and 100% to 11% comparisons, whether we included all results or only those shown as filled symbols). b. (same as Figure 4a in paper). Mutual information I in units of bits/spike. All symbols as defined in (a). The information rate in bits/spike was also significantly (p<0.01) higher during high-contrast stimuli than during either medium- or low-contrast stimuli (either when comparing all symbols or only filled symbols). c. Coding efficiency, defined as mutual information I in units of bits/s divided by total response entropy Htotal in units of bits/s (see Methods). Symbols as in (a). The decrease from high to medium or high to low contrast was significant (p<0.001 when comparing filled symbols of high and medium contrast, filled symbols of high and low contrast, or open symbols of high and medium contrast; p<0.02 when comparing open symbols of high and low contrast). d. Each symbol (○) represents the ratio of information rates (in bits/s) for a particular pair of contrasts (as indicated on the x-axis) for a single cell. The bars are the average across cells. e. The ratio of information rates in bits/spike for different contrast pairs, symbols as in (d). f. The ratio of coding efficiencies for different contrast pairs, symbols as in (d).
