Segregation of object and background motion in the retina
Citation: Bence P. Ölveczky, Stephen A. Baccus, Markus Meister (2003/05/22) Segregation of object and background motion in the retina. Nature (RSS)
DOI (original publisher): 10.1038/nature01652
Semantic Scholar (metadata): 10.1038/nature01652
Sci-Hub (fulltext): 10.1038/nature01652
Internet Archive Scholar (search for fulltext): Segregation of object and background motion in the retina
Tagged: Neuroscience (RSS) vision (RSS), retina (RSS), background (RSS), object discrimination (RSS)
This study tries to determine where the process of distinguishing local motion within a scene begins en route to the brain. This is complicated by the fact that the eye is generally in constant motion across a scene. This distinction is an essential part of detecting moving objects. Responses of different cells in the retina are studied for receptivity to this motion, and a model of retinal circuitry is proposed that explains observed effects through nonlinear pooling over interneurons.
This sort of motion detection is suppressed by global image motion - it depends on separating object from background motion. A mechanism for this is proposed, and a plausible fast-responding biological process described.
Goals and Methods
During fixation of the eye on an object, the retina still drifts over half a degree of retinal angle, at a speed of around 0.5 degrees/s. Any visual processing happens in the background of this drift. Other animals show similar drift, including salamanders and rabbits (used in this experiment). This motion has not received much attention even though a) it seems to make detecting object motion within a scene computationally hard and b) it feels effortless to humans. This study aims to find an easy way for the eye to segregate object from image motion within the retina, before the images from both eyes merge.
The authors study the spike trains of ganglion cells in isolated retinas of salamanders and rabbits. In both cases, scanning the whole retina while showing various images identified a number of ganglions that were highly selective for object motion within a scene and not at all for overall scene motion. The authors refer to these cells as object motion sensitive (OMS) cells, though they include several functional cell types. In salamanders, these are classified as ON, Weak OFF, and Fast OFF cells; in rabbits they are classified as ON-OFF Direction Selective, ON Brisk Transient, OFF Brisk Transient, and Local Edge Detectors.
Visual images are projected from a computer monitor onto the photoreceptor layer, at a mean intensity of 8mW/m. The spatiotemporal receptive fields of the retinas are mapped through reverse correlation to a flickering black-and-white checkerboard.
By studying these cells in more detail, mechanisms for suppression of coherent motion detection are considered and evaluated, along with models for selectivity for object motion. Underlying neural circuits are suggested to account for this selectivity, for both single and multiple images.
Analogies are drawn between the species studied and others known to have similar cells in their retinas.
Results and Analysis
The retina is clearly observed to sense differential motion via OMS cells, including up to 20% of the retinal cells. They are excited by motion near the center of the field.
While neurons can be suppressed by peripheral motion, this is found to arrive in brief pulses of 100ms or less, when from a similar local feature as would trigger OMS sensing in the center field. This does not account for the general inhibition of background motion. By looking specifically for interneurons that might mediate this suppression, a large amacrine cell is found that responds to coherent jitter with sharp depolarization aligning with the excitation of OMS cells.
By comparing the amacrine cell membrane potential to the time of a ganglion cell spike, an estimate of the precise function of inhibition was made. These cells are polyaxona, with the size and reach to carry out this role
Selectivity to object motion is independent of the pattern that is moving, and to the speed of jitter. And the OMS cells seem to respond to very fine elements of a scene, suggesting they may work via nonlinear pooling of the rectified response of many different subunits, similar to Y-ganglia in cats.
The similar distribution of magnocellular ganglia in primate retinas, known to have nonlinear spatial summation as well, suggests they may have similar OMS properties.
Finally, this mechanism would explain well-known motion illusions such as the Ouchi illusion, which occurs predominantly with images that have horizontal edges in the center field and vertical edges in the periphery, or vice-versa: since these perceptions are processed separately.
Theoretical and Practical Relevance
This analysis suggests that many species have cells that are specially sensitive to object motion, related to those known to carry out nonlinear spatial summation. It presents an analysis that could be carried out in a variety of other settings to probe motion tracking and object-background distinction.
It also identifies a class of illusion that this visual system explains.