Lightness constancy refers to the ability to perceive an object as having the same surface reflectance regardless of variations in illumination. The essential problem of lightness constancy lies in that the light that reaches our eyes from an object is always a product of the surface reflectance of the object and the intensity of the illumination. MaryÕs white dress in the shade and SusanÕs gray dress (of the same material and style) in the sun may well reflect exactly the same light to our eyes, and yet MaryÕs dress remains white, and SusanÕs grey.
The problem of lightness constancy is considered mathematically unsolvable without any prior constraint. During the past 30 years, remarkable progress has been made in understanding the mechanisms for lightness constancy (Gilchrist et al, 1999, for a review). It is generally agreed that, in the natural viewing environment, certain illumination cues are often present and serve as computational constraints to narrow down the possible sets of solutions. Various illumination cues have been proposed and mathematically modeled (Malony, 1999; 2002). Two kinds of cues are particularly crucial in achieving lightness constancy.
The first is a local luminance ratio cue, which is best known through WallachÕs (1948) classic demonstration. He showed that the perceived lightness of a disk embedded in a higher luminance surround is determined by the surround to disk luminance ratio. Assuming the illumination is homogeneous, changes in illumination can change the absolute luminance of a surface region quite dramatically, but the relative luminance ratios among adjacent regions often stay constant. The fact that relative luminance ratio is robust across change in illumination forms an intelligible cue for lightness constancy.
The second is a white anchoring cue, which is often referred to as the Ōhighest luminance anchoring ruleĶ. Computing local luminance ratios alone cannot give rise to the perception that we see different patches of grey as each having an absolute lightness value. Our visual system applies a simple anchoring heuristic: It assumes that the region with the highest luminance is white and then assigns lightness values to the rest of the visual scene in relation to white (Wallach, 1948; Land & McCann, 1971; McCann, 1989,1992). The availability of a white anchoring cue has been shown to have a dramatic effect on the lightness judgments in adults; lightness constancy tends to break down when subjects are tested in a complete dark room with no white anchoring cue (Li & Gilchrist, 1993; Cataliotti & Gilchrist, 1995).
Partly inspired by DannemillerÕs work (1987, 1989) showing evidences for color constancy in 4-month olds, and partly inspired by my own work on lightness induction showing evidences for WallachÕs ratio rule in 4-month olds, I initiated a series of studies on lightness constancy. My dissertation is a study of lightness constancy in 4-month olds, with a focus on the role of the aforementioned cues in achieving lightness constancy.
My approach to this question took two basic stages: Firstly, I demonstrated that 4-month-old infants show lightness constancy tested in the most favorable full cue condition. Secondly, I systematically removed one cue at a time to see if infantsÕ lightness constancy breaks down. By comparing the results between the full- and the reduced-cue conditions, the importance of each cue can be revealed. Comparable adult experiments were also included. The adult performances served as guidance to interpret infantsÕ data.
Stage 1: Testing infantsÕ lightness constancy in a full cue condition. To provide the most favorable condition to achieve lightness constancy in young infants, I used real illuminations and real paper surfaces patterned as smiley faces to attract infantsÕ attention. To make both a white anchoring cue and a local luminance ratio cue available, Experiment 1 had a white background (providing a white anchoring point cue) and stayed constant through out the experiment (providing a consistent local luminance ratio cue).
Stage 2: Testing infantsÕ lightness constancy in a reduced cue condition. Two other parallel experiments were included. To test the importance of a white anchoring cue, Experimental 2 had a constant black background in which a white anchoring point cue is absent but a local luminance ratio cue is available. To test the importance of a local luminance ratio cue, Experiment 3 had a change of background (from white to light grey) between the study and the test displays with a constant white frame in which a white anchoring cue is present but a local luminance ratio cue is unreliable.
The FNP technique was used. The key was to see if infants perceive a surface reflectance seen in one illumination as familiar under another illumination. The apparatus was a larger testing chamber with a front open window for infants to view the stimuli. The stimuli were real paper surfaces of different reflectance (roughly 60% (light grey) and 20% (dark grey)). The background reflectance was either 90% (white) Experiment 1, 5% (black) in Experiment 2, or 30% (mid-grey) in Experiment 3. Two incandescent illumination levels differing by about a factor of 3 were used. There were two parallel conditions in each experiment: a basic novelty condition, or the control condition in which the illumination was held constant between the study and the test phases, and a constancy condition in which the illumination was changed. In the study phase of each trial of the constancy condition, infants were exposed to two identical smiley faces under one illumination. In the test phase, the illumination was either increased or decreased by about a factor of 3. The infant was presented one smiley face that has the same reflectance but with a novel luminance and another smiley face with a novel reflectance but the same luminance. If infants have lightness constancy, they were expected to prefer the face with the novel reflectance regardless of changes in illumination.
The results were very interesting. Firstly, infants in all three experiments showed significant novelty preferences to a change in surface reflectacne when the illumination was constant. Second, in Experiment 1 (white background) infants showed significant novelty responses to the face with a novel reflectance, regardless of changes in illumination. This behavior indicates that the reflectance-matched face is seen as familiar across changes in illumination, indicating the presence of lightness constancy when both an anchoring cue and a luminance ratio cue are present. Third, in Experiment 2 (black background) infants showed no preference to either face, as if they both appeared novel to the infants. This behavior indicates a lack of constancy without a white anchoring cue. Forth, in Experiment 3 (change of background) some infants showed no preference to either face, and a few infants showed a reverse preference for the face with the same reflectance but with a novel luminance ratio. This behavior indicates an even stronger reduction in constancy when a local luminance ratio cue was unreliable. Pilot adult data showed a very similar pattern like infantsÕ.