I don't know too much about computer vision, but from what I do know it seems like sometimes there is an effort to make each stage of results intuitive to humans. In other words, algorithms which describe images in terms of length, width, color, scaling, etc are favored. I think its more likely that the visual representation of objects in our heads is much more of an abstract mish-mash of visual features that make no intuitive sense to the conscious mind. One technique of feature extraction used in pattern recognition is called principle components analysis (PCA), where a bunch of dimensions of input are transformed so that the axes are rotated and ordered to maximize explanatory power, and then the least useful axes are just dropped. This technique allows useless features to be ignored in the future.
Something like this might explain the babies described behavior. They recognize complete objects due to the transformation of all the object's features into a recognizable form. However, they aren't as good at tracking changes in descriptive features because features alone are not interesting - its only the combinations of features into objects that interests them. This is almost totally irrelevant to the class, and probably doesn't make any sense, but I thought I'd post something in case anyone checks this everyday and was getting disappointed in not seeing anything.
Posted by mill1991 at October 10, 2004 7:33 PM
I'm trying to respond to this comment on babies recognizing objects and such. There's a lot said here, Tim, but it reminds me of the Child Psychology class I'm taking, where we learn about 'object segregation,' whereupon a child acquires a visual representation of an object because it has visual input that are coherent, whereas the surrounding environment of the object can change.
As for how a child or anyone recognizes a 'chair' as such might have to do with rules of analogy the person forms regarding this object. Let's say a person sees their first chair, and they notice that it is different from its environment because it gets moved (for example) but stays together as a coherent unit. In addition to these visual properties, the person might notice how other objects interact with it. Another person sits on it, for instance. The child might then see a second chair and recognize it to be one by how it interacts with other objects (e.g. that people sit on it).
I don't know if I'm really addressing the spirit of your comment. I have, however, been thinking a lot about the role of words to maintain concepts in our minds (f or whichever). Perhaps words can be treated as memory devices, a bridge between the conscious mind and our long term memory (with all its associations). A word might be a label that "points to" all other associations with an object/situation/etc. we've encountered before with this word and thus bring to the surface the analogous properties we have already associated with the object. To give an example, when a child is told that a certain object is a "chair," the child might think: Oh, then, I suppose I can sit on it.
Posted by: James at October 15, 2004 2:44 PM