Depth recovery - a low-level operation

In a human visual system, it is believed that depth recovery doesn't in fact need the existence of high level information. Depth data can be derived from artificial images, notably random dot stereograms, which offer no high level cues to the viewer/computer implying that scene recognition occurs after the images have been stereoscopically fused. Humans recover relative depth quite easily, but fail in recovering absolute values. The human visual system provides subjective depth perception which provides the observer with an approximation of the spatial layout of the viewed scene. Computers, on the other hand, have difficulty in recovering both relative and absolute values.

In order to experience the human facility for the extraction of relative depth, try holding a finger up in front of you and slowly blinking your eyes alternately. In each view you will notice your finger shifting by a certain amount. Moving your finger away from your face reduces the shift whilst as it nears your face the shift becomes quite large. The shift is called disparity and is related to the distance between the viewer and the feature being viewed. In fact, if the fixation point of the two viewpoints is at infinity, then disparity is inversely proportional to depth. By comparing the relative disparities of corresponding features in two images a depth recovery process can be initiated leading to some form of three-dimensional reconstruction of the viewed scene.