Abstract
Most theories of cue combination in 3-D shape perception postulate a modular organisation of the visual system. According to these theories, several independent modules estimate 3-D properties from different sources of information. The outputs of these modules are then combined in a statistically optimal fashion by a maximum-likelihood estimator (MLE). In this empirical investigation, we studied whether the MLE model of cue integration can predict how stereo and shading information are combined. The observers' task was to place a stereo-viewed probe dot on a spherical surface that in three different conditions could be specified by (i) stereo shading, (ii) stereo-viewed random dots, and (iii) both cues combined. The MLE rule of cue integration predicts that the standard deviation (SD) of the observers settings in the 'combined cues' condition is smaller than each of the SDs in the 'stereo shading only' and 'random-dots only' conditions. Moreover, according to the hypothesis of independent modules, the predicted 'combined-cues' SD represents a lower bound of observers' uncertainty. Surprisingly, the results of this investigation show that the SD in the 'combined-cues' condition is well below (between 20 and 30 less than) the SD predicted by the MLE rule. These findings therefore indicate that shading and disparity information are not processed by independent modules.
