Differences between the actual scene model and the image model for computation of visual depth information of early vision

摘要

In this paper, we introduced two viewing modes, "Scene Mode" and "Picture Mode", for early visual depth perception depending on the dimensions of the object being viewed. The essential difference between these two modes of visual depth perception is still unclear. We discuss the basic methods of introducing a three-dimensional Cartesian system into a plane to express the depth information of an image, estimate the loss of depth information caused by this approach, and provide an analysis of the important role of providing depth information based on size constancy and vanishing point in the two viewing modes. We studied the problem of how the retina and visual cortex separate the plenoptic (all-optical) function, which is the input representation of vision, by neural computing in scene mode. We also studied the problem of how to extract information about the position and angle of light beams in the light field, and then determined the output representation of the visual depth perception. In the absence of any stereoscopic cues, such as texture, gradient, shade, shadow, color, occlusion, and binocular disparity, we compare the main differences of visual depth perception between scene mode and picture mode using a cube being viewed and its line drawing, which respectively represent the two modes.