It can see things you can't; that's a fact,
But its brain has to choose
What to keep, what to loose
And the key shapes it needs to extract.
Trying to find out what a very young baby actually sees would seem to be an almost impossible job, but a recent Scientific American article summarizing research by a team led by Jiale Yang of Chuo University, Japan, has produced some interesting results.
If you show a young baby a picture it will spend some time looking at it - but if you show the same picture twice it will spend less time on the second presentation - suggesting that it recognizes the picture as one it has seen before..If you show it picture which differ in detail it is possible to assess what kinds of differences it can detect.
In effect the young baby's eye is a bit like a camera which can see all the pixels, and can recognize similar patterns but cannot interpret them. However over the months it starts to recognize common shapes, textures, colours,etc.and, for example, recognize the family dog when seen from any angle because of the texture and colour of its coat. By the age of perhaps 8 months its brain in taking the visual images seen by the eye and data-compressing them into a range of visual concepts, which in turn can be interpreted as concepts representing objects in the real world. This ability to recognize objects from any angle, in different situations and different lighting conditions is known as perceptual constancy. As a result our eyes, once we have passed this learning stage "see" the high level concepts and not the pixels - so that if we look at two images of the same dog from different viewpoints we would recognize them as the same dog, while a very young child would see the two pictures as two different sets of pixels.
One of the results of this processing of the visual data is that we mentally "see" an image made of high level concepts rather than individual pixels and this is the basis of all optical illusions.
A similar effect happen with hearing - as a young baby learns that the important noises are the phonemes in its parents voices and other vocal noises can be ignored - which can mean they may not recognize phonemes used by different cultures using different languages.
This is very relevant to my work on the evolution of human intelligence, which organizes concepts in a recursive fashion, as it provides a framework where the lowest level "concepts" are the pixels received by the eye. These are reduced to slightly higher level concepts of colour, movement, edges of objects, etc. Several levels higher and the image is abstracted further to start recognizing real world objects. Because the process is recursive (look for patterns of concepts at one level and created a new concept at a higher level) it provides a framework where potentially a brain could handle very sophisticated concepts - and humans have taken this process several levels higher than other animals.