|
|
|
 |
Asymmetries in contrast polarity processing in young human infants
Abstract
Luminance increments and decrements of equal magnitude are processed asymmetrically in the adult visual system. At detection threshold, decrements are slightly easier to detect than increments. At suprathreshold contrast levels decrements appear to have more contrast than increments when both differ from the background by the same absolute amount. Two experiments are reported with 3.5-month-old human infants examining the processing of luminance increments and decrements. Using two different methods to measure the relative salience of positive and negative polarity high contrast bars, we found consistent evidence that dark bars appeared more salient to infants than light bars when both differed from the background by the same absolute amount. The asymmetry may be explained by noting that when luminance increments and decrements have the same Weber contrast, the decrements will have greater Michelson contrast. Perceived contrast in adults follows Michelson contrast more closely than Weber contrast, and a similar metric may characterize the relations between negative and positive contrasts in young human infants.
History
Received May 10, 2001; published November 8, 2001
Citation
Dannemiller, J. L. & Stephens, B. R. (2001). Asymmetries in contrast polarity processing in young human infants.
Journal of Vision, 1(2):5, 112-125,
Keywords
infant vision, contrast sensitivity, visual development
Features in a visual image differ from their surrounds either by being more (positive contrast) or less (negative contrast) intense than those surrounds. These features may also differ in color, but here we consider only luminance contrast differences. Our purpose here is to examine how positive and negative contrast polarities are processed by the visual system early in postnatal development. Although there have been numerous studies of contrast detection and discrimination in human infants (see below), little systematic work has examined the question of how contrast polarity is processed in the early postnatal visual system. There is evidence that positive and negative contrast polarities are not processed symmetrically in adult contrast perception (see Legge & Kersten, 1983 for a review of different kinds of evidence that support this statement). For example, sensitivity to monochromatic luminance decrements is slightly higher than sensitivity to the corresponding increments when measured in terms of the absolute difference between the target and surround (Boynton, Ikeda, & Stiles, 1964). The same result has been found with broadband stimuli (Krauskopf, 1980), although the decrement advantage is not always consistent (Cohn & Lasley, 1975). Detection thresholds for dark bars tend to be slightly lower by approximately 9% than detection thresholds for light bars across a wide range of durations and sizes (Legge & Kersten, 1983). This asymmetry at threshold in the treatment of positive and negative polarities also exists at suprathreshold levels. Bars presented as luminance increments appear to have the same contrast as bars with luminance decrements when their Michelson contrasts are equal (Burkhardt, Gottesman, Kersten, & Legge, 1984). When a central disk is presented in a surround, the brightness difference between the two appears greater if the central disk is dimmer than the surround than vice versa (Heinemann, 1955). It is important in understanding these effects to define the contrast metrics that are used. Here we consider two such metrics: 1) Weber contrast, and 2) Michelson contrast. Equal Weber contrasts represent equal absolute magnitudes of change from the background luminance. Equal Michelson contrasts represent equal percentage changes. For example, consider the situation in which dark bars have a luminance of 95 (arbitrary units) and the background has a luminance of 100. The absolute difference between the luminances of the bars and the backgrounds is 5 units. The background represents a 5.26% (5/95) change from the dark bars. To produce light bars with the same Weber contrast it is only necessary to add the absolute difference to the background to yield a luminance of 105. To produce light bars with the same Michelson contrast as the dark bars, it is necessary to increment the light bars by the same percentage over the background by which the background was incremented over the dark bars. This yields a luminance of 105.26 for light bars having the same Michelson contrast as the dark bars. Equivalently, Michelson contrast represents equal logarithmic differences [eg, ln(100/95) = ln(105.26/100)] rather than equal absolute differences. These two contrast metrics can be formally defined 1. CMichelson = (Lmax - Lmin)/(Lmax + Lmin) 2. CWeber = ΔL/L In the case of Michelson contrast, Lmax and Lmin represent the luminances of the bar and background with Lmax representing the region (bar or background) with the greater luminance. In the case of Weber contrast, L represents the background luminance and ΔL represents the difference between the bar and background luminances. The difference between these two metrics is small when the contrasts are low, but as the contrast increases, the difference can be substantial. For example, with 50% Weber contrast from a background of 100, the dark bars would have luminances of 50, and the light bars would have luminances of 150. To produce light bars with the same Michelson contrast as these dark bars would require that the light bars be set to a luminance of 200. The difference in the luminance of the light bars in this case between 150 (equal Weber contrast) and 200 (equal Michelson contrast) is substantial. The basis for this asymmetry over Weber contrast and the corresponding symmetry over Michelson contrast may be the result of retinal physiology (Legge & Kersten, 1983). The argument is that responses early in the visual cascade (eg, photoreceptors) are proportional to the log of intensity over a range around the adapting luminance. A consequence of such a log transformation is the preservation of relative responses to object-background reflectance across a range of light levels. One of the transformations that is approximately equivalent to a log transformation is the Michelson contrast transformation (see Legge & Kersten, 1983, p. 480). Opposite polarity bars that have equal Michelson contrasts will produce approximately equal responses at the first stage in the visual pathway. There is some evidence that luminance decrements are more detectable than increments for 2-month-old infants. Both Peeples and Teller (1975; see footnote 7) and Teller, Peeples, and Sekel (1978, pp. 44) report asymmetries in the discrimination of increments and decrements. Such an asymmetry is evident in the data from individual infants in these studies as well as in group averages. For example, subject Katrina from Peeples and Teller (1975) shows a slightly larger "null zone" on the positive side of the physical brightness match than on the negative side. Subjects Dina and Andrea from Teller et al (1978) show similar asymmetries in favor of decrements near threshold. There is a curious sense in which these threshold asymmetries are even more pronounced for young infants than for adults. The group average psychometric function for increments and decrements in Teller et al (1978) Figure 5 reveals this clearly. The increments and decrements around the physical luminance match are plotted on a log scale. This means that even if the positive and negative limbs of the psychometric function were symmetric on such a plot, the increments would actually differ by greater absolute amounts from the backgrounds than the comparable decrements, implying that decrements were more detectable than increments. Instead, these psychometric functions are asymmetric even when the stimulus intensity is plotted as log relative luminance. This implies that these infants show an exaggerated response to decrements relative to increments when compared to adults. The logarithmic scale equates the Michelson contrasts of the increments and decrements, but infants continue to show greater sensitivity to decrements than to increments near threshold. There is one other study in the literature showing a similar exaggerated sensitivity to decrements relative to increments in young human infants. Banks and Stephens (1982) tested contrast sensitivity for rectangular wave gratings differing in duty cycle. At the two extreme duty cycles that they employed, the patterns looked like thin dark bars on a bright background or thin light bars on a dark background. Thresholds were measured using the Michelson contrast metric. The thresholds for these extreme duty cycles should have been equivalent when measured in terms of Michelson contrast. Instead, infants 8 to 10 weeks of age were about 50% more sensitive to the dark bars on the light background than vice versa. This is consistent with the results from Peeples and Teller (1975) and Teller et al (1978) that showed greater sensitivity to decrements over increments even when Michelson or a logarithmic definition of contrast was used. There is another potential explanation for higher sensitivity to decrements than to increments at Weber contrast threshold. Poisson variability in light would lead to a higher signal-to-noise ratio (S/N) for decrements than for increments when both differ from the background by the same absolute amount. The variance in the number of photons reflected from a surface or emitted by a radiant source over a fixed period of time is proportional to the mean level. The S/N ratio will be greater in the region of the decrement because the absolute amount of light in that region will be less. Detecting the same absolute difference in luminance from the background should be easier in the region of the decrement because the variance in that region over repeated trials will be less than in the region of the increment.1
With the exception of these threshold data from the 2-month-old infants, there are no other data at suprathreshold contrast levels that tell us how the early postnatal visual system treats increments and decrements. The goal of this study was to test the hypothesis that suprathreshold luminance decrements are more perceptually compelling or salient to young infants than are equal absolute magnitude increments.2 To test this hypothesis, we used two different methods for examining the influence of contrast polarity on infants' initial fixation behavior.
Method This experiment used a standard first-look method.3 Briefly, infants were presented with a display having 28 small bars. All or most (see details below) of the dark (negative contrast) bars appeared on one side of the display, and all or most of light (positive contrast) bars appeared on the other side of the display. An observer then recorded to which side of the display the infant looked first when these bars appeared in the visual field from an otherwise uniform background. The light and dark bars had the same Weber contrasts, which gave the dark bars greater Michelson contrast. The hypothesis is that most of the first looks should be directed to the side with the dark bars if Michelson contrasts better characterize suprathreshold contrast perception than Weber contrast in young infants. Participants Infants were recruited from birth announcements in a local newspaper. Twelve infants provided complete data for this experiment. The average age of these 12 infants was 98.5 days (range, 92-104 days). All infants who were tested contributed complete data. Apparatus and Stimuli The stimuli were presented on a large monitor running at 60 Hz in a noninterlaced frame mode. At the 50-cm viewing distance, the stimulus field was 40 degrees horizontally X 31 degrees vertically. The background color of the stimulus field was yellow (x = .503; y = .439) as were all of the bars. The bars were 5 degrees vertically by 0.75 degrees horizontally.
The luminance levels for the bars and the back-ground are shown in Table 1 (Experiment 1). The positive contrast bars (increments) and the negative contrast bars (decrements) differed from the background by the same absolute luminance. Thus, their Weber contrasts (100%) were equal in magnitude but opposite in sign. Michelson contrasts for the two polarities are also shown in Table 1. Notice in Table 1 that when equal luminance increments and decrements are used, the Weber contrasts are equal, but the Michelson contrasts are unequal with the negative polarity bars having more contrast (100%) than the positive polarity bars (33%). The display was situated at the infant�s eye level in a matte black wall. The infant sat in an infant seat facing the display. To the infant�s right of the display, there was a peephole that an observer used to watch the infant�s eye and head movements and to make on-line judgments. The observer used a button box interfaced to a computer to start the trials and to register right and left judgments. Prior to the start of each trial, a small blue flashing bar appeared in the center of the screen to attract the infant�s attention. The observer could also use sound-making toys centered at the infant's midline but not visible to the infant to induce orienting before the start of a trial. All of the bars appeared on the display simultaneously after the centering bar was removed. The same practiced observer was used with all of the infants. Design and Procedure This experiment used a first-look methodology. The adult who was observing the infant judged the direction of the infant's first look or the most prominent direction of regard after the bars appeared on the display. This adult observer was "blind" to the display characteristics on all trials. The latencies to make these judgments were on the order of 1.5 to 2 seconds (see below). The experimental variables were manipulated within subjects. Forty trials were presented to each infant.4 There were three different trial types: 14 dark bars versus 14 light bars (20 trials) 14 dark bars versus 13 light bars plus 1 dark bar (10 trials) 14 light bars versus 13 dark bars plus 1 light bar (10 trials) Trial type 1 is a standard preference condition. Trial types 2 and 3 are "pop-out" conditions. One bar on one side of the display differed in polarity from the 13 other bars surrounding it. This "odd" bar always appeared centered vertically and horizontally on one half of the display across trials with the side randomly determined but balanced within conditions. If this local feature contrast played a role in determining first looks, then one might expect to see differences from the preferences shown in the standard condition (trial type 1). The side with the dark and light bars was counterbalanced within each trial type. Trials were run in blocks. Each block contained one trial each of types 2 and 3 and two trials of type 1. The order of the trial types within each block was randomized. Ten such blocks were presented to each infant. The 14 bars on each side were distributed between 14 imaginary columns that divided the horizontal extent of the display into 14 equal segments. Two bars appeared in each column. The vertical positions of the static bars in the columns were random with the constraint that two bars could not overlap and the whole of a bar had to be visible. This produced a display with 28 bars more or less randomly distributed across its extent. The goal was to simulate a situation in which the infant had multiple potential attentional targets within this portion of his/her visual field. Schematic examples of the displays are shown in Figure 1.
Results and Discussion The data from this experiment are shown in Figure 2. The percentages of first looks to the side with most of the dark bars are shown for each infant. The mean percentages of first looks to the side with most of the dark bars in the two pop-out conditions did not differ significantly (type 2 M = 73.3%; type 3 M = 75.0%), so these data (20 trials) were averaged for each infant and presented as the open symbols in Figure 2. The closed symbols represent the standard 14 dark versus 14 light bars trial type. There was a strong tendency for infants to direct their first looks toward the side with most of the dark bars. There was no significant difference between the standard preference condition and the mean from the pop-out conditions. In both cases, approximately 71%-74% of first looks were directed toward the side of the display with most of the dark bars. Notice also that every infant in both conditions always directed more first looks toward the side with the dark bars than toward the side with the light bars. In both conditions, the mean percentage of first looks was significantly above 50%. The observer made these directional judgments on average in 1.62 seconds (SD = 0.18 seconds). Across the three trial types, the average judgment times ranged from 1.58 seconds (SD = 0.25 seconds) for trial type 3 to 1.69 seconds (SD = 0.19 seconds) for trial type 2. The average judgment time did not differ significantly between any of the conditions. These data show that when equal luminance increments and decrements are pitted against each other at high contrast, infants tend to look first at the decrements. This result is compatible with the hypothesis that Michelson contrast qualitatively predicts preferences at 3.5 months of age better than does Weber contrast. The Weber contrasts of the light and dark bars were the same, but these infants all showed a reliable preference for the dark bars.5 Informally, several adults were asked to choose which set of bars appeared to have more contrast. All chose the dark bars. The difference in the apparent contrast between the dark bars and the background versus the light bars and the background was substantial for these contrast levels.
Method Our goal in the second experiment was to use a different method to test the same hypothesis. The results of Experiment 1 showed that when all or all but one of the dark bars on the display were segregated on one side of the display, there was a strong preference to attend initially to the side with the dark bars. This preference can be extended in three ways. First, a less complete segregation of the dark and light bars would be expected to yield less robust preferences and at the same time possibly generalize the results to conditions with less extreme spatial segregations of the light and dark elements. Second, the preference for dark bars can be put into competition with another stimulus that generally attracts attention at this age: motion. We asked whether the spatial distribution of the dark and light bars would exert any detectable effect on the frequency with which infants attended to the side of the display with a single moving stimulus - a motion "singleton" in the language of visual search. Third, the preference shown in Experiment 1 was observed when the Weber contrasts of the bars were equal leading the dark bars to have greater Michelson contrast. This preference should disappear when the Michelson contrasts are equalized by increasing the Weber contrast of the light bars relative to the Weber contrasts of the dark bars. These three changes were used to test the generality of the preference for dark bars shown in Experiment 1 under more heterogeneous stimulus conditions. Briefly, the method used in this experiment is a variant of one that we have used previously (eg, Dannemiller, 1998). A small oscillating bar appears with 27 static bars. In the current case, half of the bars on the display were increments over the background and half were decrements. These bars were distributed spatially more or less randomly across the display with two important constraints: 1) on each trial there were always 14 bars on each half of the display, 2) across trials the spatial distribution of the increment bars and decrement bars was unbalanced with a ratio of 11:3. If the two polarities appear to have equivalent contrast, then this spatial distribution manipulation will have no effect on initial orienting to the moving bar. In contrast, if the decrement bars appear to have more contrast and are more perceptually salient, then orienting to the moving bar may depend on the location of the 11 decrement bars relative to the moving bar. When most of these bars appear on the side of the display contralateral to the moving bar, they should compete with the moving bar to capture the infant's attention, and orienting to the moving bar should occur less frequently than when these more salient bars appear on the same side of the display as the moving bar. The extent of the competition will depend, of course, on the salience of the moving bar. In our past research with infants at this age, we obtained reliable competition effects with the bar oscillating at 1.2 or 2.4 Hz through an amplitude of 2.0 degrees (leftmost position to rightmost position). We have observed the pattern of results consistent with competition repeatedly in previous work using differences in color contrast and differences in luminance contrast without polarity differences (Dannemiller, 1998; Dannemiller, 2000; Ross & Dannemiller, 1999; Dannemiller, in press). In other words, when the spatial distribution of two differentially salient classes of static bars is manipulated, competition induces differences in how readily infants orient to the moving bar. A significant spatial distribution effect is then used to infer differences in the perceived color or luminance contrast of the two classes of static bars. Participants Infants were recruited from birth announcements in a local newspaper. Forty-eight infants provided complete data for this experiment. The average age of these 48 infants was 98.0 days (range, 90-105 days). Data from another 8 infants were excluded for the following reasons: excessive crying, fussiness or inattentiveness (n = 7), and birth complication severe enough to necessitate a stay in the intensive care unit (n = 1). Apparatus and Stimuli The apparatus and stimuli were identical to those used in Experiment 1 with several minor changes described below. The luminance values and contrasts for the two conditions in this experiment are shown in Table 1 under Experiment 2. Each infant received 40 trials. These 40 trials were distributed equally between four different stimulus conditions. The four stimulus conditions in each group comprised a 2 x 2 within-subject factorial design. One factor was the polarity of the single moving bar; it was dark on half the trials and light on the other half of the trials. The other factor was the location relative to the moving bar of most of the light bars. In the ipsilateral condition, the side with the moving bar had 11 light bars and 3 dark bars. The other side of these displays had the complementary ratio of light to dark bars 3:11. In the contralateral condition, the side opposite the moving bar had 11 light bars and 3 dark bars. The other side of this
display had the complementary ratio of light to dark bars 3:11. The terms ipsilateral and contralateral always refer to the location of most (11/14) of the light bars relative to the moving bar. Thus, there were always 14 dark and 14 light bars on the display, but they were more unevenly distributed than they had been in Experiment 1. All of the bars appeared on the display simultaneously after the centering bar was removed. The same practiced observer was used with all of the infants. Examples of these displays are shown in Figure 3. The moving bar always appeared in one of two locations on each trial: in the middle of the display vertically and either 10 degrees to the right or left of the center of the display. The 27 static bars could appear anywhere on the display with the following constraints. Thirteen of the static bars appeared on the same half of the display as the moving bar. The remaining 14 static bars appeared on the half of the display opposite the moving target. Thus, a total of 28 bars appeared on the display on every trial and were evenly divided between the two sides of the display. The spatial distribution of the 14 bars on each side followed the same rules as those used in Experiment 1 with two bars appearing randomly in each imaginary column. Design and Procedure Data were collected using the Forced-Choice Preferential Looking Technique (FPL; Teller, 1979). The adult who was observing the infant made a forced choice on each trial about the location of the moving bar. This adult observer was "blind" to the trial type and to the location of the moving bar on each trial. The computer provided the observer with feedback about the correctness of this judgment after every trial in the form of a brief audible beep. The FPL observer was instructed to make these judgments as quickly as possible while maintaining reasonably good accuracy because we were interested in initial orienting or alternatively in the dominant direction of regard in the seconds immediately following the onset of the stimulus bars. It is more common with the FPL technique to allow the FPL observer to wait indefinitely on each trial until enough evidence has accumulated to make a forced choice judgment. This version of the FPL technique differed because the observer made a speeded judgment. The latencies to make these judgments were on the order of 1.5 to 2 seconds (see below), so we are confident that this measure gives us information about orienting during the initial second or two after these stimuli appeared. The basis for these judgments rests on the direction of the first look on most trials, similar to the behavior used in Experiment 1. 
There were two groups in Experiment 2 with 24 infants randomly assigned to each of these groups. In the Weber-equal group, the Weber contrasts of the positive and negative polarity bars were equal at 100%. The Michelson contrast of the dark bars in this condition (100%) was greater than the Michelson contrast of the light bars (33%). In the Michelson-equal group, the Michelson contrasts of both the dark and light bars were equal at 33%. The Weber contrast of the positive polarity bars (100%) was greater than the Weber contrast of the negative polarity bars (50%) in this Michelson-equal group. Our strategy was to run identical experimental protocols with either the Weber contrasts or the Michelson contrasts of the dark and light bars equal. The pattern of results should reveal which of these two metrics more closely characterizes the responses of young infants to luminance contrasts of opposite polarity. Hypotheses In the Weber-equal group, we expected to see a lower percentage of correct judgments in the ipsilateral condition than in the contralateral condition. Recall that in the ipsilateral condition most of the light bars appear on the same side as the moving target, leaving most of the higher salience dark bars on the side opposite the moving target. It is this condition in which we would expect to see the most competition with the moving bar. Additionally, although the Weber contrasts of the dark and light bars are equal in this group, the Michelson contrasts of the dark bars are much greater than the Michelson contrasts of the light bars (see Table 1, Experiment 2, Weber-equal). In the Michelson-equal group of this experiment, we expected to see no effects of the spatial distribution of the two polarities because the effective contrast of the two polarities should have been equal in this condition. Notice that this is a null prediction, but it is accompanied by a non-null prediction for the other group, so it is essentially a prediction of an interaction of the group variable with the spatial distribution variable.6 An auxiliary hypothesis is that overall percentages of correct judgments may be higher with the moving dark bar than with the moving light bar but only in the Weber-equal condition where the Michelson contrasts differ. Results and Discussion The percentage of correct judgments again served as the dependent measure in a mixed ANOVA. The between-subject factor was group (Weber-equal versus Michelson-equal), and the within-subject factors were the moving bar polarity (positive versus negative) and the spatial distribution variable (ipsilateral versus contralateral) depending on which hypothesis was being evaluated. The major hypothesis predicted an interaction between the spatial distribution variable and the grouping variable. The spatial distribution variable should exert an effect in the Weber-equal group, but there should be no effect of this variable in the Michelson-equal group. The means from all conditions for both groups are shown in Figure 4. The spatial distribution X group interaction was significant, F(1, 46) = 6.10, P = .017. The spatial distribution factor exerted a significant effect on the percentages of correct judgments in the Weber-equal group (Figure 4 left panel) but not in the Michelson-equal group (Figure 4 right panel). The effect in the Weber-equal group was in the predicted direction. The percentage of correct judgments was lower when most of the positive polarity static bars were on the same side as the moving target (ipsilateral in Figure 4, M = 62.3%) than when they were on the side opposite the moving target (contralateral in Figure 4, M =74.8%). Switching the location of most of the negative/positive polarity bars with respect to the moving bar modulated the percentage of correct judgments by an average of 12%. The negative polarity static bars in the Weber-equal condition had substantially more Michelson contrast (100%) than the positive polarity bars in this condition (33%). In the Michelson-equal group, the mean percentages of correct judgments did not depend on the spatial distribution of the two static bar types (M = 64.8% and 66.0% for ipsilateral and contralateral, respectively). The major hypothesis was supported. The data from individual infants also showed this spatial distribution effect. In the Weber-equal group, 17 infants showed less attention to the moving bar when most of the negative polarity bars were contralateral to the moving target than when they were ipsilateral; four infants showed the opposite pattern; and three infants showed no difference. A sign test showed this to be a nonrandom result in the predicted direction (P = .007). In contrast, in the Michelson-equal group, 12 infants showed better detection when most of the negative polarity static bars were ipsilateral to the moving bar than when these bars were contralateral to the moving bar; nine showed the opposite pattern; and three showed no difference (P = .664). The mean judgment time in the Weber-equal group (M = 1.720 sec) was slightly less than the mean judgment time in the Michelson-equal group (M = 1.767 sec). This difference of 47 milliseconds was not significant, t(46) = 0.75, P = .459. The auxiliary hypothesis discussed above led to the prediction of a target polarity X group interaction. The polarity of the moving target should exert an effect on the overall percentage of correct judgments, but only in the Weber-equal group because the Michelson contrasts of the two bar types are substantially different (Table 1) in that group. The negative polarity moving bars should have produced higher percentages of correct judgments than the positive polarity targets. The pattern in the data supported the auxiliary hypothesis. There was an interaction between the polarity of the moving target and the grouping variable, F(1,46) = 4.06, P = .05). In the Weber-equal group, the mean percentage of correct judgments (M = 71.3%) with a moving decrement was higher than the mean with a moving increment (M = 65.8%). This difference was not observed in the Michelson-equal group. In the Michelson-equal group, infants oriented toward the moving decrements slightly but not significantly less often than they did toward the moving increments (M = 64.4% versus 66.5%, respectively). Infants more readily oriented to the moving bar when it was dark than when it was light but only when their Michelson contrasts were unequal. The pattern of results across the two groups in this experiment followed exactly from the hypothesis that the effective contrast of both the static and the moving bars more closely reflects the Michelson contrast metric than the Weber contrast metric. When the Michelson contrasts of the two polarities were equal at 33%, the detection rates for both polarity moving bars were equal, and the spatial distribution of the two polarities had no effect on the percentage of correct judgments despite large differences in the Weber contrasts of the light and dark bars in this condition. When the Michelson contrasts of the two polarities were unequal (100% for negative versus 33% for positive), the negative polarity static bars were more effective than the positive polarity bars in competing with the moving bar for the infant's attention despite the fact that the Weber contrasts of the light and dark bars were equal in this condition. Finally, we also note that prior to this experiment we ran the Weber-equal condition with lower contrasts for the light and dark bars (Weber contrasts 52%; Michelson contrasts 21% and 35% for the increments and decrements, respectively). The pattern of results was in the predicted direction, but the differences between the ipsilateral and contralateral conditions were close to null. Thus, the greater salience of the dark bars that we observed in the Weber-equal group in Experiment 2 may only hold with very high contrasts.
The major results of these experiments can be summarized succinctly. For 3.5-month-old infants, the perceptual effectiveness of luminance decrements is greater than the effectiveness of luminance increments when both differ from the background luminance by the same absolute amount. This asymmetry held in several ways. First, infants directed more of their first looks toward the side of a display that had all or most of the dark bars than toward the side with all or most of the light bars when there was no moving bar present in the display (Experiment 1). Second, a moving dark bar drew attention more readily than a moving light bar despite both having the same Weber contrasts (Experiment 2, auxiliary hypothesis). Third, and similarly, static dark bars competed more effectively for attention with a moving bar than static light bars despite both having the same Weber contrasts. These results can be explained by assuming that perceived contrast at 3.5 months of age as in adults more closely follows Michelson contrast than Weber contrast. Comparison of the Weber and the Michelson contrasts for each condition shown in Table 1 shows that the pattern of results observed in both experiments is more consistent with a Michelson than with a Weber contrast metric. As such, these results extend previous contrast detection results for 2-month-old infants showing this same asymmetry in favor of decrements (Peeples & Teller, 1975; Teller et al, 1978; Banks & Stephens, 1982) to suprathreshold contrast levels. The presumptive reason for this asymmetry is that photoreceptor responses at this early age are proportional to the log of intensity relative to the background intensity as they are in adults. Michelson contrast is one transformation that approximates this logarithmic relation (Legge & Kersten, 1983). It must be noted that the predicted pattern of results in the Weber-equal condition of Experiment 2 did not hold when we used a smaller contrast difference and lower average contrast in a pilot experiment (21% versus 35% for the Michelson contrasts of the increments and decrements, respectively). This could have occurred because the contrast difference between the two polarities in the pilot experiment may have been near the threshold of contrast discrimination for infants at this age (Stephens & Banks, 1987). In a previous study using only light bars of differing contrasts, we found mixed evidence for a spatial distribution effect (Ross & Dannemiller, 1999). We found a small, but significant spatial distribution effect in the predicted direction using Michelson contrasts of 20% versus 33%; when most of the higher contrast bars appeared contralaterally to the moving bar, the percentage of correct judgments was less than when most of these bars appeared ipsilaterally to the moving bar. This effect was in the predicted direction but was not statistically significant when luminance increments with Michelson contrasts of 50% versus 66% were used. Thus, in the pilot experiment for the current paper, our failure to find a significant spatial distribution effect may reflect the unreliable nature of contrast discrimination at this age with smaller contrast differences. The contrast vision literature with adults offers several plausible suggestions for why luminance decrements may be more detectable or salient for young infants. Kontsevich and Tyler (1999) have shown that the detection of contrast changes by adults is different for low and high spatial frequency periodic patterns. At low spatial frequencies, detection of contrast changes was mediated by detecting luminance decrements in the dark bars. In contrast, at higher spatial frequencies, detection was based primarily on noticing the bars that became brighter. Given the low-pass character of early contrast sensitivity at this age (Banks & Salapatek, 1981; Fiorentini, Pirchio, & Spinelli, 1983 ; Gwiazda, Bauer, Thorn, & Held, 1997), it may be that the dark bars are simply more visible to the infants, and hence draw their attention more readily than the light bars. Korth, Rix, and Sembritzki (1992), using the pattern electroretinogram (PERG), have shown in adults that the onset of dark bars produces a spatial-frequency tuned, pattern response very similar to the PERG evoked by a full pattern-onset-offset stimulus consisting of both light and dark bars. In contrast, the onset of exclusively light bars on the same background produces a response that is not tuned with spatial frequency and resembles the response to a full-field luminance increase. Again, the dark bars may simply be more visible as a pattern to these infants than the light bars on the opposite side of the display. It is important to point out one limitation of these conclusions. The greater effectiveness of decrements relative to increments at 3.5 months of age may depend on the relative strengths of transient responses to these stimuli. Infants saw these bars appear suddenly from a uniform field. Thus, the effectiveness with which the static bars competed for attention may have depended on luminance decrements producing stronger transient onset responses than the onset transients produced by luminance increments. In a standard preferential looking study, infants are allowed to look at the display for an extended period of time, and their preferences are measured by how long they look at each side of the display rather than by the side of the display at which they first look. Would the same pattern of results hold if longer inspection times were used? The data do not permit us to speculate on this because the display times were short--approximately 1.75 seconds. It is important to remember that the asymmetry that we have found should only be generalized to situations with abrupt onsets of spatially limited luminance increments and decrements. There is another potential explanation for the results that we obtained with both of these methods. The behavior that was used most often in both of these experiments was the direction of the infant's first look after the bars appeared on the display. Is it possible that latency differences of neural elements in the visual pathways that respond to positive versus negative contrasts could give a slight temporal advantage to dark bars in terms of recruiting attention? There is some evidence from retinal recording in other species that the responses of neural elements that signal offsets (decrements) begin slightly earlier than the responses of elements that signal increments. Burkhardt and Fahey (1999) have shown that amacrine cells in the salamander retina exhibit shorter latencies to respond to negative contrast flashes than to positive contrast flashes. The difference is approximately 20 to 45 milliseconds. In contrast to this physiological result, Burkhardt, Gottesman, and Keenan (1987) showed, using simple reaction time measurements in human adults, that the reaction time evoked by maximum negative contrast (decrements) was always longer than the reaction time evoked by maximum positive contrast. In fact, when reaction time is used as the measure to derive equivalent positive and negative contrasts, the data fall along a line described by equal Weber contrast rather than equal Michelson contrast. Differences in the response latencies of neural elements in the visual system that respond selectively to local luminance increments versus decrements do not appear to explain the psychophysical results with adults, but at present we know of no comparable data from infants at this age that would allow us to determine the status of this hypothesis for explaining the present results. In addition to differences in behavioral response measures, complex and/or meaningful stimulus configurations may challenge the application of our results to other preferences in human infants. Although not designed to identify contrast polarity effects per se, some stimuli employed in infant face preference experiments may be easily approached by predicting a consistent preference for stimulus pairs that differ only in terms of contrast polarity. For example, Dannemiller and Stephens (1988) showed that 3-month-old infants prefer a "normal" schematic face over its contrast-reversed version. Is this preference due to contrast polarity processing asymmetries? We think not. Six-week-old infants show no such preference, and neither age group exhibited a preference between two "abstract" control stimuli that also differed only in contrast polarity. Predictions based only on contrast polarity metrics cannot account for all of these preferences. Therefore, other mechanisms (eg, top-down processes) are required to account for the observed preferences that employ complex or meaningful stimuli. The paradigm that we used in Experiment 2 depends on putative competition effects. What process(es) might be responsible for these effects? In our past work, we have used a signal detection model to explain these effects (Dannemiller, 1998). This model assumes that each of the bars on the display leads to an internal signal to saccade to that location. We assume that these signals are perturbed by internal, random noise. The infant then looks on each trial to the side of the display with the maximum internal signal. This is usually the side with the moving bar, but on a nontrivial proportion of trials because of the random noise, the largest signal may arise from one of the static bars on the display. If this maximum occurs on the side contralateral to the moving bar, then it leads the FPL observer to make an incorrect judgment because the correctness of the judgment is defined by the location of the moving bar. This model is similar to others that have been proposed in the visual search literature with adults (eg, Palmer, Ames, & Lindsey, 1993). The spatial distribution effect is predicted by this model if there is a large enough difference in the mean internal responses to the two classes of static bars. In the current experiment, this model explains the spatial distribution effect as arising from the larger mean internal response to decrements than to increments. An alternative plausible model assumes that on some percentage of trials, the moving bar leads to perceptual pop-out as a moving singleton typically does in visual search studies with adults. When pop-out occurs, the infant orients reliably to the moving stimulus. On the complementary percentage of trials, this pop-out effect is absent either because the infant is not attending or possibly because of high internal noise in the channels responsible for signaling the moving bar. What determines orienting on these remaining trials? A reasonable guess is that the maximum response model described above now describes the decision to look right or left based on the side with the static bar that produced the largest internal response. Standard signal detection theory with multiple noise samples captures behavior on these trials when motion pop-out fails to occur. The first look then tends to be directed toward the side with more of the dark bars in a manner similar to that shown when all of the bars were static in Experiment 1. This latter model qualitatively captures the set of results from both experiments reasonably well. One question that remains with regard to the alternative or pop-out model is why motion pop-out doesn't occur on every trial. In addition to keeping in mind that it is impossible to instruct infants to attend to a "target" stimulus, several other factors suggest themselves as answers to this question. Past psychophysical research with infants has shown that even with very strong stimuli, percentages of correct judgments in a two-alternative forced-choice task may not asymptote at 100% (Teller, Mar, & Preston, 1992). This is usually attributed to attentional fluctuations on the part of the infant. It could also reflect noise in the FPL observer's judgments. At present, there is no way to decide among these sources for the less than perfect pop-out percentage implied by the alternative model and the data from Experiment 2. Our purpose in discussing this issue was simply to point out that this assumption of fluctuating attention is not uncommon in research with infants. It can be handled formally by assuming that the probabilities associated with the infant's correct responses are a weighted mixture of two psychometric functions; on k percent of the trials when the infant is attending, the response is determined by one psychometric function, and on (100 � k)% of the trials, the response is random or determined by a different psychometric function with a shallower slope (see Viemeister & Schlauch, 1992 for a formal model). Finally, we return to the evidence cited in the "Introduction" that young infants may actually exhibit exaggerated sensitivity at threshold to decrements relative to increments (Peeples & Teller, 1975; Teller et al, 1978; Banks & Stephens, 1982). Recall that even when the contrasts of increments and decrements were measured using a Michelson or a logarithmic metric, young infants showed enhanced sensitivity to the decrements. Our data do not speak directly to this issue because the stimuli that we used were well above threshold. We did observe, however, in Experiment 2, that when Michelson contrast was made equal, the asymmetric response to light versus dark bars disappeared. This could reflect the fact that the method is not sensitive enough to detect the asymmetry at suprathreshold levels. It could also mean that the exaggerated asymmetry near threshold disappears well above threshold. One possible explanation of the exaggerated asymmetry at threshold involves a greater compressive nonlinearity on the intensity versus response function in infants compared to adults. If the response to increments above the background luminance level saturated with increases in contrast at lower levels in infants than in adults, then infants might show greater relative sensitivity to decrements relative to increments. In other words, if the response to a spatial increment that raised the luminance to twice the surrounding level was weaker than the response to a spatial decrement that dropped the luminance to half its value in the surrounding area, then the increment could appear to be less visible or detectable. A strongly compressive nonlinearity on an intensity versus response function perhaps at the level of the photoreceptors could produce this type of exaggerated asymmetric response to contrast. In summary, when large luminance increments and decrements of equal absolute magnitude appear simultaneously in the infant's visual field, the decrements appear to be more salient to 3.5-month-old infants. Decrements compete with a moving bar for the 3.5-month-old infant�s attention more effectively than increments. This asymmetry is similar to that observed in adults both at detection threshold and at suprathreshold levels when perceived contrast is measured. The asymmetry is predictable from the hypothesis that perceived contrast for infants follows Michelson contrast more closely than Weber contrast. This may reflect the responses of mechanisms early in the visual pathway that respond to the logarithm of intensity relative to the background intensity (Legge & Kersten, 1983).
This research was supported by National Institute of Child Health and Human Development Grant R01 HD32927 to J.L.D. We thank Jacqueline Roessler for observing the infants, Manya Qadir for scheduling the infants, and Daniel Replogle for all of the computer programming.
We thank Martin S. Banks for this observation. We thank Davida Teller for suggesting the polarity manipulation in a conversation at ARVO 2000. We actually conducted the two experiments in this paper in the reverse order, but because this experiment used a simpler method, we present it first. Two additional trial types were presented to each infant with 10 trials presented for each type. Displays with all dark bars and displays with all light bars were presented to check for infant/observer side biases. No such biases were evident, with the mean left versus right percentages being 53% and 50% for these trial types. We do not discuss these trials further. We checked for two potential stimulus artifacts that could have indicated the difference between the two sides of the display to the adult observer. First, the corneal reflection available to the observer was not sufficiently detailed to support discrimination of the two sides of the display within the 1.5 to 2.0 seconds limit of these judgments. Second, when we replaced the infant's face with a sheet of white paper to check for luminance gradients, the observer was unable to determine which side of the display contained the dark versus light bars from looking at the white paper. We presented two other trial types to each infant in both of these conditions, but these trials are not discussed here. In addition to the trials with mixed polarity bars on the display, each infant also saw 10 trials with all increments (including the moving target) and 10 trials with all decrements. These two trials types are useful for testing process models of competition and for checking on the consistency of parameter estimates in such models, but they do not bear directly on the issue of whether or not the two polarities are differentially salient.
Banks, M.S., & Salapatek, P. (1981). Infant pattern vision: A new approach based on the contrast sensitivity function. Journal of Experimental Child Psychology, 31, 1-45. [PubMed] Banks, M. S., & Stephens, B. R. (1982). The contrast sensitivity of human infants to gratings differing in duty cycle. Vision Research, 22, 739-744. [PubMed] Boynton, R.M., Ikeda, M., & Stiles, W.S. (1964). Interactions among chromatic mechanisms as inferred from positive and negative increment thresholds. Vision Research, 4, 87-117. [PubMed] Burkhardt, D.A., Gottesman, J., Kersten, D., & Legge, G.E. (1984). Symmetry and constancy in the perception of negative and positive luminance contrast. Journal of the Optical Society of America A Optics and Image Science, 1, 309-316. [PubMed] Burkhardt, D.A., Gottesman, J., & Keenan, R.M. (1987). Sensory latency and reaction time: Dependence on contrast polarity and early linearity in human vision. Journal of the Optical Society of America A Optics and Image Science, 4, 530-539. [PubMed] Burkhardt, D.A., & Fahey, P.K. (1999). Contrast rectification and distributed encoding by ON-OFF amacrine cells in the retina. Journal of Neurophysiology, 82, 1676-1688. [PubMed] Cohn, T.E., & Lasley, D.J. (1975). Spatial summation of foveal increments and decrements. Vision Research, 15, 389-399. [PubMed] Dannemiller, J.L. (2000). Competition in early exogenous orienting between 7 and 21 weeks. Journal of Experimental Child Psychology, 76, 253-274. [PubMed] Dannemiller, J.L. Relative color contrast drives competition in early exogenous orienting. Infancy. In press. [PubMed] Dannemiller, J.L., & Stephens, B.R. (1988). A critical test of infant pattern preference models. Child Development, 59, 210-216. [PubMed] Fiorentini, A., Pirchio, M., & Spinelli, D. (1983). Development of retinal and cortical responses to pattern reversal in infants: A selective review. Behavioural Brain Research, 10, 99-106. Gwiazda, J., Bauer, J., Thorn, F., & Held, R. (1997). Development of spatial contrast sensitivity from infancy to adulthood - psychophysical data. Optometry and Vision Science, 74, 785-789. [PubMed] Heinemann, E.G. (1955). Simultaneous brightness induction as a function of inducing- and test-field luminances. Journal of Experimental Psychology, 50, 89-96. [PubMed] Kontsevich, L.I., & Tyler, C.W. (1999). Distraction of attention and the slope of the psychometric function. Journal of the Optical Society of America A, 16, 217-222. [PubMed] Korth, M., Rix, R., & Sembritzki, O. (1992). The different contributions of local luminance decreases and increases to the pattern electroretinogram (PERG). Vision Research, 32, 229-237. Krauskopf, J. (1980). Discrimination and detection of changes in luminance. Vision Research, 20, 671-677. [PubMed] Legge, G.E., & Kersten, D. (1983). Light and dark bars: Contrast discrimination. Vision Research, 23, 473-483. [PubMed] Palmer, J., Ames, C.T., & Lindsey, D.T. (1993). Measuring the effect of attention on simple visual search. Journal of Experimental Psychology: Human Perception and Performance, 19, 108-130. [PubMed] Peeples, D.R., & Teller, D.Y. (1975). Color vision and brightness discrimination in two-month-old human infants. Science, 189, 1102-1103. [PubMed] Ross, S.M., & Dannemiller, J.L. (1999). Color contrast, luminance contrast and competition within exogenous orienting in 3.5-month-old infants. Infant Behavior & Development, 22, 383-404. [PubMed] Stephens, B.R., & Banks, M.S. (1987). Contrast discrimination in human infants. Journal of Experimental Psychology: Human Perception and Performance, 13, 558-565. [PubMed] Teller, D.Y. (1979). The forced-choice preferential looking procedure: A psychophysical technique for use with human infants. Infant Behavior and Development, 2, 135-153. Teller, D.Y., Peeples, D.R., & Sekel, M. (1978). Discrimination of chromatic from white light by two-month-old human infants. Vision Research, 18, 41-48. [PubMed] Teller, D.Y., Mar, C., & Preston, K.L. (1992). Statistical properties of 500-trial infant psychometric functions. In L.A. Werner, & E.W. Rubel (Eds.), Developmental Psychoacoustics (pp. 211-227). Washington, DC: American Psychological Association. Viemeister, N.F., & Schlauch, R.S. (1992). Issues in infant psychoacoustics. In: L.A. Werner, & E.�W. Rubel (Eds.), Developmental Psychoacoustics (pp. 191-209). Washington, DC: American Psychological Association. [PubMed]
|
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
![[home]](/images/article/home.gif)
![[e-mail]](/images/article/mail.gif)
 |
 |
 |
 |
jov |  |
 |
 |
 |