There is general agreement that edges play a central role in determining the colour and lightness appearance of the surfaces through similar filling-in mechanisms. However, the way through which their influence is performed is still unclear. Two different theories have been put forward to explain the filling-in completion phenomenon.
One theory ─ we will address it as the ‘Isomorphic filling-in theory’ according to the definition of Von der Heydt, Friedman et al. (2003) ─ postulates that perception is based on an image representation held in a two dimensional array of neurons, typically arranged retinotopically, in which colour signals spread in all directions except across borders formed by contour activity. The process is thought to be analogous to physical diffusion, with contours acting as diffusion barriers for the colour and brightness signals. An alternative hypothesis is that image information is transformed at the cortical level into an oriented feature representation. Form and colour would be derived at a subsequent stage, not as the result of an isomorphic filling-in process, but as an attribute of an object or proto-object. This theory is called the symbolic filling-in theory.
According to the isomorphic filling-in theory, colour is represented by the activity of cells whose receptive fields point at the surface, but it is assumed that these cells receive additional activation through horizontal connections that keeps their activity level high despite mechanisms of lateral inhibition tending to suppress surface activity and despite the transient nature of the afferent signals. The lateral activation comes from receptive fields at contrast borders. These signals are strong because receptive fields are exposed to contrast, and reliable because the border produces continuous light modulation even during fixation, due to small residual eye movements. In the alternative symbolic hypothesis, there is no spreading of activity, but all the information would be carried by the relevant features, that would be tagged with information on contrast polarity, colour and lightness of the surfaces they enclose. Despite the many attempts to verify the two different models by psychophysical and physiological experiments, the mechanisms of colour and lightness filling-in are still debated.
Though intriguing, these results cannot be easily generalized to similar phenomena, such as the filling-in of illusory contours or the filling-in through artificial scotomata or adapted edges (such as in the Troxler’s effect). All these phenomena are indeed similar, and probably rely on similar neural circuitries but they are not identical. For instance, an obvious difference between filling-in across the blind spot and filling-in of occluded edges is that filling-in across the blind spot is modal (i.e. you literally see the filled-in section), while filling-in across occluders is amodal. Filling-in across the blind spot was found to be different also from filling-in across cortical scotomata in two patients examined by Ramachandran (Ramachandran 1992; Ramachandran, Gregory et al. 1993). In these subjects, some features filled in the scotoma faster than others, and in some circumstances filling-in took some seconds before it was completed (while filling-in across the blind spot is immediate). Together these data suggested that mechanisms for the filling-in of colours, motion and texture can be dissociated and may correspond to processes in higher-order areas that are specialized for these attributes.
Paradiso and Nakayama (1991) performed an experiment to verify this hypothesis. They presented a large disk of uniform brightness on a black background. The stimulus was briefly flashed and, after a variable stimulus offset asynchrony, a masking stimulus was presented. The mask consisted of a circle on a black background with the masking contours positioned within the boundaries of the large uniform disk. This experiment is grounded on the assumption that filling-in consists of a spreading of neural activity from the boundaries of luminance and through the surfaces, that is stopped when another luminance-contrast border is reached (this is proposed by many models of brightness perception, see for example Walls 1954, Gerrits and Vendrik 1970, Cohen and Grossberg 1984), and that the process takes some time to be completed.
Subjects were asked to match the brightness at the centre of the disk with a palette of grey scales. When the delay between target and mask presentation was long enough, the mask had no effect on the apparent brightness of the stimulus, but for stimulus offset asynchronies of 50-100ms, the surface of the disk inside the masking annulus appeared unfilled. Moreover, the minimum target-mask delay at which the masking was effective increased with target size, suggesting that there would be a spreading phenomenon and that the farther the features delimiting a region, the more time is necessary for the filling-in to be completed. These results are supported also by further experiments on temporal limits of brightness induction in simultaneous contrast (De Valois, Webster et al. 1986; Rossi and Paradiso 1996; Rossi, Rittenhouse et al. 1996), as well as by a similar experiment performed by Motoyoshi (1999) on filling-in of texture.
Electro-physiological recordings in retinal ganglion cells, LGN and primary visual cortex showed that neurons of these areas responded to luminance modulation within the receptive field even in the absence of contrast borders.
In a second condition, a uniform grey patch was placed on the receptive field (extending 3- 5 degrees beyond the receptive field boundary on either side), and two flanking patches modulated sinusoidally in time from dark to light. With such stimuli, the brightness of the central patch appears to modulate, despite the absence of luminance change. In this condition, cat retinal ganglion cells and lateral geniculate nucleus cells, having their receptive fields centred in the uniform grey patch, did not respond; on the other hand, primary visual cortex neurons were modulated by luminance changes far outside their receptive fields. Together, these results suggest that neurons in the retina and LGN are responsive to luminance modulation, but their response does not correlate with perceived brightness. On the other hand, striate neurons responded to stimulus conditions producing changes in brightness in the area corresponding to the receptive field.
The behaviour of primary visual cortex neurons seems to be in agreement with the one hypothesized by an isomorphic filling-in theory in that they both respond to luminance of the surfaces also in the absence of borders, and their activity is modulated by that of edges far outside the receptive field. Moreover, when the temporal frequency of luminance modulation in the surrounding patches exceeded a threshold value, the induced response disappeared, suggesting that it was the result of a spreading of activity, taking a finite time to happen, likely explainable in the context of isomorphic filling-in.
Dennet and Kinsbourne (Dennet 1992; Dennet and Kinsbourne 1992) opposed to the idea that an active filling-in process would take place in our brain on philosophical grounds. They argued that such an idea would be the result of the false belief that in our brain there is a spectator, a sort of homunculus similar to ourselves, needing a filled-in image representation. From a scientific viewpoint, Dennet’s homunculus may correspond to higher-order scene representation or decision-making mechanisms. The question is whether or not such mechanisms need a filled in, gap free representation of the image to function optimally (Ramachandran 2003).
The symbolic filling-in theory postulates that such a “homunculus” need not exist, and that image information is transformed at the cortical level into an oriented feature representation. Surface form and colour are not coded at this stage, but would be derived only at a symbolic level of representation, as attributes of objects or proto-objects.
The authors recorded the activity of surface- and edge-cells (cells whose receptive fields pointed either to the filled-in surface or to the border between the disk and the ring) in the visual cortices V1 and V2 while the monkey was performing the filling-in task. The activity of surface-cells correlated with the physical stimulus change in both areas V1 and V2, but not with the perceived colour change induced by filling-in. The activity of edge-cells followed the stimulus contrast when the disk colour changed physically; when the colours were constant, the edge signals also decayed, but more slowly. Together, these data are incompatible with the isomorphic filling-in theory, which assumes that colour signals spread from the borders into uniform regions.