Colour helps to solve the binocular matching problem

Abstract

The spatial differences between the two retinal images, called binocular disparities, can be used to recover the three-dimensional (3D) aspects of a scene. The computation of disparity depends upon the correct identification of corresponding features in the two images. Understanding what image features are used by the brain to solve this binocular matching problem is an important issue in research on stereoscopic vision. The role of colour in binocular vision is controversial and it has been argued that colour is ineffective in achieving binocular vision. In the current experiment subjects were required to indicate the amount of perceived depth. The stimulus consisted of an array of fronto-parallel bars uniformly distributed in a constant sized volume. We studied the perceived depth in those 3D stimuli by manipulating both colour (monochrome, trichrome) and luminance (congruent, incongruent). Our results demonstrate that the amount of perceived depth was influenced by colour, indicating that the visual system uses colour to achieve binocular matching. Physiological data have revealed cortical cells in macaque V2 that are tuned both to binocular disparity and to colour. We suggest that one of the functional roles of these cells may be to help solve the binocular matching problem.

Figure 1. Stimuli and experimental procedure

A 2D stimulus presented to each eye (A), in which the disparities between the images determine depth of the perceived 3D stimulus (B), upon which observers had to indicate the observed depth-to-width ratio using a vertical slider that could be manipulated with the computer mouse (C).

Figure 2. Top view of the geometry of the binocular matching problem

The intersections of the visual lines indicate possible matches. The filled discs indicate the correct matches, the open discs are the false matches. A represents the possible matches in a monochrome stimulus. B shows how introduction of a stimulus feature such as colour (different lines) reduces the number of possible matches.

Figure 3. The statistical linear mixed effect models and their fit to the perceived depths

In model 1, the null hypothesis, all variance is assumed to be error variance around the mean, not modulated by any of the factors; therefore there is no parameter on the x-axis. In model 2, the separate observers are allowed different means, but all individual variance is still assumed to be error variance around the individual means. In model 3 an effect of complexity is shown; as the stimuli become more complex, less depth is perceived. This effect is assumed to be the same size in all individuals. In model 4A it is assumed that the effect of complexity is modulated differently in each observer. Each subsequent model has a better fit to the data (last graph) compared to the previous model.

Figure 4. The perceived depths (A) and predicted perceived depths (B) for the 20 different conditions for the two different colour information categories (monochrome versus trichrome)

The five different number of bars are represented by different symbols; 6(▪), 24(♦), 48(▴), 72(*) and 96 (○). The two luminance conditions (congruent and incongruent) are represented by dotted or continuous lines. A shows effects of complexity (F_1,7= 16.992, P < 0.01), colour (trichrome > monochrome, F_1,966= 13.024, P < 0.001) and luminance (congruent > incongruent, F_1,966= 8.771, P < 0.01) on the perceived depth.