Visual processing by nerves, no hand-waving words as to how

[Billy T]

I first comment on the problem of perception of 3D world "out there" constructed with 2D retinal information. All the processes that convert that information in to a highly compressed (nearly by a factor of 100) data stream in the optic nerve are well understood. The most of the information via the LGN comes to the cortical called V1. That which goes elsewhere permits what is known as Blind sight (but I will not speak much of it as it mainly gives minor assistance to vision, with little of no perception - Monkeys with V1 removed, can still pick up a nut from the floor, without groping for it. A larger fraction of their retinal data than in humans goes directly to the superior colliculus that does this. In non-humans that area of the brain is usually called the Tectum.)

Many who concern them selves with visual processing gloss over the first problem in V1. Namely how does the continuous 2D field of neural excitation get parsed into discrete objects. My suggestion for how is discussed in detail in the Ref. 1 of the link below - a paper I published back in 1994. I was studding what is known about how nerves of the brain interact. Here is a discussion of that: http://toritris.weebly.com/visual-system-3-v1---line-detectors.html
Hubel & Weisel got the noble prize for the discovery of "line detectors" in V1, which BTW they are not but all call these orientation sensitive cell "line detectors." Those cells are more complex than H&W realized.

The stimulus that H&W presented to the money with indwelling micro-electrodes in V1 was field of uniformly spaced parallel black on white lines on screen they could rotate. (So no need to control where the money was looking). If you do a 2D spatial Fourier (Gabor function more correctly) on that grid pattern stimulus, Yes it has strongly spiked orientation, but also a strongly spiked spatial frequency. Their cells are, I am nearly sure are detecting both. I.e. those H&W cells are actually doing a transform of the visual field to "Fourier like" space. See figures A, B, C &D here:
http://mri-q.com/uploads/3/2/7/4/3274160/fourier_transform_and_ph_enc.pdf

The last, D, is a well known photo of president Lincoln in both normal form and in the Fourier transform - both contain exactly the same information; however, the FT version does not change if the original image from the real world is shifted a foot left or even if made half as large (but the scale of the FT does change in this second case,) BUT NOT THE RELATIVE WEIGHT of the components. - That makes identification of object in FT space independent of factors, like location, that don't change what the object is. Why I believe, and experiments suggest, the brain operates on the FT version of the field of view at later stages, say in the temporal lobes where object identification is done.

But still they need first to be parsed into object from the continuous field of neural stimulation in V1. In my next post is how, I believe that is done (compressed version of first part of Ref 1 of link below).

here is more related to ALL perception I hope reader will find interesting:
http://www.sciforums.com/showpost.php?p=905778&postcount=66 there I explain and justify my RTS view of perception with focus on showing genuine free will is not necessarily inconsistent with the natural laws that control the firing of every nerve in your body. Then see:

http://www.sciforums.com/threads/wh...e-will-an-illusion.104623/page-5#post-2644660 and posts 84,86 & 94 where I clarify my POV more.

Next post here will return to the more narrow question as to how do neurons parse continuous field of stimulation in V1 into separate "objects" for later brain stages to identify.

 

[Billy T]

This post returns to the more narrow question as to how do neurons parse a continuous field of stimulation in V1 into separate "objects" for later brain stages to identify. Currently it is copy from a prior post and some typed drawings that may be improved later by edit.

In post 1, I tell about H&W's mis-named "line detector" cells, here after, called LDCs. In this post I tell how they interact to parse object from a continuous stimulation field, but note lot of the work was already done in the retina. - I.e. by and large, only contrast information is sent to the brain via the optic nerve. Retinal regions with intensity and color uniformity are not. That is how the system works with the neural fibers of the optic nerve be less than 2% as numerous as the retinal cells. - huge data compression in the retina gives V1 a less difficult task.

My education was in physics, especially applied physics. Thus I knew that a network of electric generators making say 60Hz power is much more frequency stable than any generator in the network is. Thus when I learned that to nearby LDCs with the same orientation sensitivity re-enforce each other's activity and mutual suppression occurs between nearby cells with orthogonal orientation sensitivity - a "light bulb" flashed in my mind. The solution to problem of parsing the V1 field of stimulation into objects, was suddenly was clear to me. Already most of the activity in V1 was in contrast boundary form, thanks to retinal data compression processing It mainly, except for sharp corners, consisted of boundary lines slowly changing in orientation.

For example, LDCs with like orientation and physically approximately "end to end" like this:
|
|
|
or like this: - - - - - -
Would fire in phase as each is stimulating the other, and just that month some researchers had noted that cells in V1 separated by a few mm were indeed synchronized and firing with a frequency of ~40Hz.

The problem came at those corners and when view of one object was partially blocked by another object which was closer to the eye. For example, the image of a black cat walking behind a white picket fence is perceived as an "integrated cat" not as a set of vertical "cat strips."

More reading of literature after H &W let me learn that there are "angle detectors" Ax, for all possible angles at all locations in V1, where "x" is some specific angle, like 64 degrees, but of course a 60 degree angle between two lines will excite it too, just not ast0ngly as the 64 degree angle would. If x = 90 degrees then I'll call Ax = C a "90 degree corner detector."

This I postulated that if the mutually re-enforcing boundary turned 90 degrees, like this:

|
|
|
C - - - - - - then the corner detector could transfer excitation (both ways) at its sensitivity angle, in this case 90 degrees. So the lowest of the vertical bars in may illustration above would get suppression from the left most horizontal bar but reinforcement from C, so would be sort of "net neural," but all of this would soon be firing at the same frequencies even if the vertical bars had mutually "locked together" initially at 30 hz and the horizontal line detectors at 40 hz. I. e. both would be "pulled" (like AC electric generators with slightly different near 60hz frequencies say 60.1 & and 59.9hz pull each other to 60hz.) to a common firing, in-phase, frequency.

Thus any set of line detectors making a "closed loop" would rapidly settle down to fire in phase with a unique frequency typically in the range of 25 to 50 hz. That was what "marked" them to be a separate, parsed object; however, returning to the cat walking behind the picket fence that idea alone would indeed suggest we should perceive the cat not as a unified object but as a set of parallel "cat strips."

I don't know if I can explain well in my next post without the set of drawing in my 1994 paper the solution to this "cat strip problem" still keeping at the neurotic level, (not at the higher Gestalt level of the "law of good continuation," which like most of cognitive is just words, not explaining how the brain actually does what it does with neurons.) but I'll try.

 

[Billy T]

In this post 3 you are told to ignore all the dots - but I will edit them from your view by coloring them white - thanks to hit from Tiassa. also when I learn how, I'll try to make alignments better.

It is an empirical fact that equally well defined closed loops that are large settle into one common in-phase firing frequency faster than smaller ones do. That is why we perceive the black cat walking behind the picket fence as one object and not as a set of verticall cat strips. Completely ignore the dots in the line drawing below - they just prevent Sciforum's computer from compressing several spaces down to only one.

......__
.....|.. ..|
----|---|----------------------------
|.................................................| Again please ignore all dots.
|.................................................|
-------------------------------------
.....|.. ..|
.....|.. ..|
.....|__|

I intend the above to look sort of like an opaque horizontal box with a smaller more narrow box crossing behind it (cental section blocked from view).
(Cut me some slack - best I can do without paint brush etc.)

The bigger box has much more mutual re enforcement occurring so say comes into synchronized oscillation as at 45 hz before the smaller vertical box can. Lets assume the vertical side wall of the little box are oscillating in phase at 35 hz. Focus your attention on the red horizontal section. The big box has claimed it as part of it boundary (has 45hz firing) before the top visible part of the little box can (gets into its 35 hz oscillation). The little box would like to be two: one on top distinct from the slightly longer one on the bottom, but it too, can not become a closed loop as the big box has claimed for its own the top part that could close this bottom part. Perhaps the incomplete bottom section initially was starting to oscillated at 38Hz.

The top 35 hz and the bottom 38hz do suppress the amplitude of the red sections of the 45 hz oscialtion, but not enough to stop them. Even though there is no light stimulating some of the vertical LDCs as there would be if the big opaque box did not exist, these LDC with vertical sensitivity are be stimulated by the vertically sensitive LDCs just adjacent to the red sections. I.e. some "no-light" activity does exist in the vertically sensitive LDCs that are needed to make the little box one object, instead of two unrelated ones. If not too far apart, these not-light activated LDCs can "bridge the gap" and get into "closed loop oscillations" - get an ID as one object.

That, at the neuronal level, IMO, explains how the Gestalt Law of "Good Continuation" is achieved by neurons.

I'll stop here, but just note that the 1994 paper also discusses how motion data (processed in V5) is also used to make objects complete and then further how these defined object get identified - for example how a non-moving back cat is distinguished from an equally large image of an old style black telephone when both (due to slightly different distance from the eye are both activating the same number of retinal cells - this too is a problem as you don't know what these masses of "black activity" in your nerves are until AFTER that problem is solved.

 

[Billy T]

This "off subject" post explains why this thread exists. Skip it if not interested.

Three prior post were replies to a new member, who had read a great deal on QM, consciousness, and what is reality, but IMO was somewhat confused by it all. None the less he was writing a book about it all and posting blocks of his text in thread he created. Here is an example - one part of one block - that prompted my reply in blue, below.

" ... the reason one may make quantum observations with classical apparatus is due to the existence of a Stoke's line of asymptotic expansion, named after Sir George Gabriel Stoke's. The fact that one can gain infinitely new values within a given magnitude already gives almost ubiquitous perspective, built upon semi-classical framework. The classical region defines within our perception of a quantum mechanical behaviour, while a large wave packet covers all classical regions. "

The spelling of "behavior" suggest either he is English or is just copying from an English writer's text. In any case the existence of the stokes lines (considerable effort required to see more than one) is not "the reason" but an example of a vibrational quantum being transferred from (or very rarely to) a photon. I have never read any writing of Stokes, and don't know if he understood he was probable the first to have evidence that energy was quantized when very small amounts were being observed. Stokes probably only understood that fluorescence could change the wave length - making it longer. I doubt it he ever observed what is now called the "anti-stokes line" where the newly created radiation has more energy (shorter wave length) than the exciting light.

http://www.encyclopedia.com/topic/Sir_George_Gabriel_Stokes.aspx
Tells me Stokes: "was a professor of mathematics at Cambridge; he served as secretary (1854–85) and as president (1885–92) of the Royal Society. His researches, done in many fields, developed the modern theory of viscous fluids, revealed the nature of fluorescence, and helped to establish the composition of chlorophyll. The important work he did on the undulatory theory of light led to publication of his Dynamical Theory of Diffraction (1849). His other publications include Light (1884) and Natural Theology (1891)."

Thus it is very probable that he did not anticipate Planck as the first to realize optical energy, at least, was quantized.

I replied to new member, The Moon, with {in part}: "... as graduate student I photographed the Stokes and Anti-Stokes lines generated in clear liquid Carbon Disulfide. The spectrograph used had a shutter plate you could slide down blocking light as it did. I moved it down a little after about a minute exposure, then several times more (7 or so? - it was ~50 years ago so I forget) small steps each time with the last exposure on the bottom edge of the glass* photo sensitive plate at least an hour long. - The anti-Stocks line (excitation energy plus one vibration quantum of the liquid more energetic) was clearly visible, at least in the last full hour exposure, (perhaps one above it too?) but others were badly over exposed.

* High resolution spectro-photo-graphs always use glass plates as they don't shirk - and distance along the lines, from known calibration lines, is how you get their wave lengths. ..."

 

[iceaura]

This post returns to the more narrow question as to how do neurons parse a continuous field of stimulation in V1 into separate "objects" for later brain stages to identify. Currently it is copy from a prior post and some typed drawings that may be improved later by edit.

For starters, what is presented to the neurons is not continuous. The eye does a lot of processing before any brain stages, early or late, register anything.

Also, there are brain injuries or disorders that disable the construction of continuity - and therefore prevent the perception of motion, among other effects. Victims see a series of still pictures - they cannot safely cross a street with traffic.

 

[Billy T]

For starters, what is presented to the neurons is not continuous. The eye does a lot of processing before any brain stages, early or late, register anything.

My continuous refer to in space not time, but yes it is continuous in same sense as digital photo graph is. Yes, a lot of the work is done in the retina. I said than in 2nd sentence of the OP, which was:
"All the processes that convert that information in to a highly compressed (nearly by a factor of 100) data stream in the optic nerve are well understood."
and soon went on to tell that mainly contrast boundaries were extracted and sent on via the optic nerve. Adding that large area with little change were not (sent on).

Also, there are brain injuries or disorders that disable the construction of continuity - and therefore prevent the perception of motion, among other effects. Victims see a series of still pictures - they cannot safely cross a street with traffic.

Yes that is true, but your speaking of continuity in time, not space. Parsing spatial continuous (for all practical purposes as in the fovea the resolution is very high) is the problem I was addressing. There is a point in each retina where the optic nerve exits that does not have photo sensitive cells but fortunately that is a different point in the visual field for each of the eyes. Consequently most don't even know they have these "blind spots" as neural processing fills them in - no hole in perception with both eyes open. If one eye only open that "filling in" also occurs, but it is then a "best guess" based on nearby stimulation of the retina of the "open eye" and can be very wrong. It is easy to demonstrate this with a thin vertical line, all green except for small red section with tiny black x in the red area.

Any text on the "blind spot" will tell you how to position that x in the blind spot so it is not seen. Do that, and your perception will be of a continuous vertical only green line. I.e. the red section along with the x is not seen and the nearby green section provides the data used to "fill in." - Why your perceive falsely a only green, continuous, vertical line. This is somewhat related to the Gestalt law of "good continuation" I discussed in post 3.

There are several visual defects that make crossing a street hazardous for some. I think you are concern with one where the nerves do not process motion, (done in visual area V5) correctly. - a temporal discontinuity in perception. Another is failure of steady depth perception. People with normal vision automatically use depth information to make what is called "size constancy." I.e. the retinal image of a man walking towards you from say 30 meters way to only 6 meters away become five times larger but the perception of his height remains essentially constant. - size constancy. This can be and usually is defective in people who were congenitally blind and later had sight restored, to a large extent because their depth perception is defective. Very rare they can gain normal depth perception, even if all other aspect of normal vision are mostly restored with time and experience.

There are some people who have a false dynamic depth perception. A parked car can sudden be perceive to "jump" at them or the converse: a car moving towards them can be perceived to go into reverse. Not very safe for them to try to cross a street. Some people have non responsive areas in V1 bind spots call scotoma.* As the same part of the visual field from both eyes falls on the cortical scotoma they have, and are very aware of, blind spots. For them an approaching car can "disappear."

* I had some contact with doctors and researchers in the Johns Hopkins "Wilmer eye clinic." One group there was working on intentionally distorting the image presented to the retina of people with well defined and known scotomas. Idea was to blow up the optical image that would fall on the scotoma and compress it into the surrounding area. I. e. No information would be lost and the subject could soon learn to "un-distort it" I.e. fill in the scotoma. I lost contact with this group so don't know their progress. They knew that even if it worked as they hoped, it would an "expensive cure" as location and shape of every cortical scotoma is unique - more of a research project to better understand the limits of "filling in" I think.