Visual stability across saccades

Look from one word to the next. Your eyes move quickly, but the page does not seem to sweep past. The brain must distinguish this self-produced image motion from something actually moving in the world. How much of that distinction comes from the images themselves, from repeated experience, or from predicting the eye movement?[1] [2] [3]

Moving your gaze between words shifts the image on the retina, yet you normally see a stable page. These rapid eye movements are called saccades. The open question is how image interactions, learned sensitivity and predictions of the movement together distinguish self-produced motion from changes in the world.

Active accounts use an efference copy: a copy of the eye-movement command predicts incoming input. Passive accounts exploit interactions between images arriving close together in time. Habituation adds history-dependent sensitivity. A recurrent neural-network model develops inhibitory remapping and image-centered coordinates when optimized for an energy-efficiency objective; these are computational results, not identified biological circuitry.[1] [2] [3]

Surrounding images can hide a brief motion signal, and repeated eye-motion pairings can make it less visible. It remains unclear when these processes suffice and when a prediction of the eye movement is needed.[1] [2] [3]

That fully connected model is not a biological demonstration of remapping necessity. Human grating effects likewise do not establish natural-scene weights. Retinally matched stimuli, manipulated exposure and selective circuit perturbations could distinguish prediction-dependent omission from visual masking and learned sensitivity.[1] [2] [3]

The answer would explain how useful changes stay visible while the disturbance from an ordinary glance largely disappears.[1] [2] [3]

Stable vision has to preserve useful detail as well as omit the disturbance from a glance. Orientation integration constrains that first requirement: information survives across views, with extra weight given to the central, sharp view in one task. Motion omission constrains the second. Explaining both would connect the continuity of what we see with the selective loss of self-produced motion, without treating these different measurements as interchangeable.[1] [2] [3]