This discussion board is for any thoughts/questions you might have on the physiology of vision and "the plenoptic function and the elements of early vision" by Adelson and Bergen

I guess this is pretty trivial, but I'm sort of confused about their use of the term "pencil." I had always taken it to mean "a one-parameter family of lines." But here they use it to refer to the set of all lines passing through a point in E3. But this is a 2-dimensional family, right (spherical coordinates, or the cartesian coordinates of a picture plane)?

I want to quote the following from the paper:

"The significance of the plenoptic function is this:
The world is made of three-dimensional objects, but these
objects do not communicate their properties directly to
an observer. Rather, the objects fill the space around them
with the pattern of light rays that constitutes the plenoptic function, and the observer takes samples from this
function. The plenoptic function serves as the sole communication link between physical objects and their corresponding retinal images. It is the intermediary between
the world and the eye."

I think that Adelson is being overly metaphysical when he says this. This plenoptic function reminds me of the ether, a stationary material medium that pervades all space, which you might have heard about during an introduction to Einstein's theory of special relativity.

Although it is beneficial to talk about a function whose dimensions correspond to basic visual properties, it might be a better idea to simply talk about embedding basic visual measurements in a vector space instead of making deeper philosophical claims regarding the existence of the plenoptic function.

According to Mathworld, a "pencil" is all lines thru a point:
http://mathworld.wolfram.com/Pencil.html
However, a "pencil of planes" is a line.

Edited by author 01-17-2006 11:42 PM
I think the main contribution of this paper is to establish a mathematically plausible model for the elements of early vision based on many psychophysical and physiological evidences. (Am I Right?)

Currently, is Adelson’s model one of very plausible hypothesizes or a proven fact? (because this paper is quite old ( Published in 1991.)) Is it true that our visual cells (I'm sorry I'm not familiar with biological terminology) work like a combination of derivative and smoothing functions?

Another fundamental question; how important is it for us (computer scientists) to follow up this kind of research (Cognitive science?) I totally agree that it has been provided us with many good ideas or insights to develop new computer vision algorithms.

As a follow-up to Gunhee's question on whether this model really coincides with what living vision cells do: As far as I know, visual cell's activation patterns do in fact at least partially correlate with several of the presented operators of the plenoptic function. Not only has there been psychophysical evidence such as presented in the papers' appendix, but there has also been more recent neuro-scientific evidence which shows similar response patterns as those of plenoptic derivative operators.

I remember one particular cognitive science talk where they actually measured life cell activation patterns of a ferret's visual cortex using microwire electrodes. It turns out that the activation pattern really represents basic visual features (such as contrast, orientation). I believe this is the paper matching the talk: http://www.bcs.rochester.edu/people/weliky...n_natural_scene.pdf

I think plenoptic functions shouldn't be considered as a model for human vision system. They do not describe how the cells work, but rather describe what they can possibly do.

Adelson was clever to find a way to organize low-level features within combination of 2 out of 5 dimensions that can be processed in the early part of the vision system. It would be interesting to watch all of the boxes in his "periodic table" get filled by future researchers. I really do appreciate his work. But as the author pointed out a few times in his paper, there are stimuli (even simple ones) which involve more than 2 dimensions. In that case, his cute 2D squares depicting a stimulus will become cubes in 3D, 4D, ... and it will no longer look cute.