Some Interesting Things About Flies..

[cozdas]

...Therefore I see no reason why an insects brain does not perform a similar combination of separate images to form a similar 3D version of what is around it.

I guessed you didn't mean a flat image too, but wanted to make sure that we are talking about same thing: a "3D model" instead of 2d flat image. I totally agree with you on your above quote.

I cannot believe for instance a dragonfly can be so good at catching insects in flight if it is only seeing a rudimentary fuzzy image?

Just like human eyes, it doesn't need to have a uniform resolution across the field of view. We humans have 1min-arc resolution at the fovea but that drops pretty quickly when you move away from the optical center.


Being able to move our eyes and head lets us to sample the scene in parts on demand and progressively build a high-res model of the surrounding. After all although we have 100 million photoreceptors we have just 1 million nerves going to the brain but we think our vision is much better than 1 megapixel, that is true because with moving parts (both the visible eye movements and very small and fast jittering eye movements) we refine the image progressively. If you try to read just 4-5 lines above while staring at this dot (.), you realize how fuzzy even our stationary visions are but it's not a big issue in practice.

After checking the web for dragon-flies (which turned out to have ~30K ommatidia) I saw that they have a similar structure where in the central part there are much more receptors per angle. from http://www.daviddarling.info/encycloped ... d_eye.html

Compared with single-aperture eyes, such as the human eye, compound eyes have poor resolution so they are not good at making out detail. On the other hand, compound eyes have a very large angle of view and the ability to detect fast movement and, in some cases, the polarization of light. Insects that can fly well, such as honey bees and flies, or that catch prey, such as dragonflies or preying mantis, have specialized zones of ommatidia. These zones are organized into a fovea area that gives acute vision. In the acute zone, the eye is flattened and the facets are larger, which allows more ommatidia to receive light from a spot and thereby achieve higher resolution.

Compound eyes generally allow only a short range of vision. For example, flies and mosquitoes can see only a few millimeters in front of them with any degree of resolution, although within this short range they can see detail that we could see only with a microscope.

Dragonflies have one of the most elaborate eyes of any insect, capable of pinpointing the motion of a small prey insect several meters away, even his while the dragonfly is traveling fast. Butterflies have color vision that is more enhanced than our own, enabling them to locate food from flowers. Honey bees can see in ultraviolet, which allows them to perceive patterns on nectar-laden flowers that are invisible to us. Many insects, including bees, can also detect polarized light, which they use in navigation.

[rjlittlefield]

cozdas, thanks for tracking down these excellent references and wordsmithing a summary explanation. Very helpful!

So for these kinds of compound eyes the overall optical system seems to be like that:
- each ommatidium corresponds to a partial cone of the entire field of view but those cones are not mutually exclusive meaning those cones from different ommatidia are overlapping.
- photoreceptors in each ommatidium corresponds to (probably mutually exclusive) fraction of the cone of that ommitidium.
- Some receptors of an ommitidium share the same view direction with some other receptors from some (but not all) other ommitidium. In other words a particular view direction is represented by some photoreceptors in few ommitidia.

it's a pretty similar to 4D light field structure that I'm familiar with in computer graphics, but unlike the light field sensors, compound eyes are not flat but curved thus low-res sub images does not cover the same FOV.

That sounds like a pretty good description to me.

The bottom line is I learned that my "one lens = one photoreceptor" knowledge was valid only for some sub-set of the compound eyes. While reading that I also remembered this thread and re-read Rik's explanation. I guess he was describing one of the other superposition eye structures.

Not quite. What I wrote was really intended as a quick and dirty description to counter the "mosaic of images" model that is exemplified by the classic Gary Larson cartoon "The last thing a fly ever sees". Gary knows better -- he's a trained biologist and to him the picture is a joke. But a lot of people (myself included, for many years) do believe that at some level of processing the fly has to deal with a collection of highly detailed images, one per ommatidium, when in reality it does not.

Each facet of the fly's eye (called an ommatidium) is like a single pixel in a digital image. It provides a signal that represents the average brightness of some small region around the fly.

Taken literally, those words would best describe a simple apposition eye, one value per ommatidium. Superposition eyes are somewhat more complicated, but on a scale of values-per-ommatidium they are still much closer to one than they are to the thousands of pixels that would be implied by Larson's cartoon or by the words "miniature cameras" that appeared in anvancy's post. For example in "Modeling the First Two Neural Layers in the Common Housefly Vision System", the authors Moya, Wilcox, and Donohoe write that:

Under each facet lens lies a set of eight photoreceptors, six peripheral cells surrounding a pair of tandemly arranged central ones. ... From each of these sets, one peripheral receptor joins five others from under adjacent facet lenses in sampling the same general region of the visual space. ... This single visual element is repeated several thousand times across each eye.

So in essence each ommatidium in this case is kinda sorta 6 pixels, with the caveat that the input image is smeared across multiple receptors. In a direct experiment using photosensitive dye, Moya et.al. show that one particular point source of light in the far field stimulates all 6 receptors in each of 5 neighboring ommatidia, and some fewer receptors in each of 6 more surrounding ommatidia, thus being seen to some extent by a total of 11 ommatidia. The two neural layers described in the paper then apparently try to process the overlapping signals so as to recover some amount of the resolution loss that would otherwise be implied by smearing a point source over a total of 42 sensing units in 11 ommatidia.

I cannot believe for instance a dragonfly can be so good at catching insects in flight if it is only seeing a rudimentary fuzzy image?

Just like human eyes, it doesn't need to have a uniform resolution across the field of view.

And just like in weaponry, it doesn't need to have very good resolution as long as it's sensitive enough to acquire and fast enough to track. Simply being able to detect the target and then hold it within a cone of several degrees as the distance drops would be more than sufficient for capture.

--Rik

Edit: fix broken link

[DaveW]

One thing touched on but largely overlooked is the brains persistence of vision, though this is often ascribed to the eyes. If we did not have this we would see a cine film as a number of still pictures, not a moving one. But surely it also applies to still images in the brain. As said, the eye scans a scene rather like the scanning lines on a TV picture but the screen shows a full image due to the persistence of the phosphors and human brain as I understand it? Although our eye is constantly scanning the scene when we stare at something it tends to remain a fixed image in our brain, not a succession of lots of individual bits of the scene.

I see no reason why that should not happen with an insects brain either. In botany parallel evolution has produced the modification of many different features to provide virtually similar results and functions, so no reason this should not occur in the animal kingdom. Whilst our eyes have sharp central vision the peripheries of our sight are also less sharp, but like a flies eye very sensitive to movement which immediately switches the sharper central vision to lock on to this.

As said before, with all science we can only hypothesise from the structures before us what other than humans see, we probably will never actually know for certain what a flies brain actually sees. After all science is littered with indisputable truths such as the sun goes around the earth and the earth is flat.

Tomorrows scientific hypotheses often replace today's. The "truth" is something that apparently fits the facts at the time and allows us to predict things. Future scientific discovery may prove it is something different that produces similar results. We can never know what are absolute truths in science, only hypotheses that seem to fit the known observations.

DaveW