SHAPE REPRESENTATION AND MULTIPLE DEPTH CUES

点击翻译按钮获取中文摘要

DESCRIPTION (Investigator's Abstract): Biological visual systems make use of multiple depth cues including occlusion, texture, perspective, motion parallax, disparity, shading, and so on. The first of three themes that guide the work proposed here is that an ideal depth observer is sensitive to the quality of information available from different depth cues in a particular scene. The second theme is that the ideal depth observer is sensitive also to the logical type of information (in a measurement-theoretic sense) available from different depth cues. The investigators will argue that depth information of different types must be promoted to a common type before being combined, and this process of promotion may explain many observed depth cue interactions. The third theme concerns the veridicality of stimuli (how closely they resemble 'real world' scenes).If an ideal depth observer is sensitive to the quality of information available from different depth cues, then the use of impoverished or distorted depth stimuli can in itself alter the course of depth processing. The experimental work proposed will make use of hardware and software capable of generating highly realistic, controllable viewing conditions involving large displays, observer motion, and accurately modeled scene and surface properties. The experimental methods used are based on perturbation analysis methods that permit analysis of a system that can potentially react to distortions and inconsistencies in stimuli. The proposed research consists of four major tasks. (1) The investigators will test the general applicability of a particular model of cue combination, the linear model, which the investigators have previously demonstrated for combinations of texture, motion and stereo. This will involve estimation of depth using additional depth cues as well as applications to other tasks including determination of absolute spatial location and 2D feature localization. (2) The investigators will examine whether the human depth combination rule is statistically robust in the sense of discounting cues that indicate depth values discrepant from other cues. (3) Exploration of the time course and limits of cue calibration for absolute depth will be carried out using rich 3D displays and an open-loop pointing task. (4) Experiments will probe the forms of depth representation used by human observers.