The delicate graphs over low-contrast, well-shot photos are so beautifully subtle. And the typographer in me loves the equally understated equation captions.
Seems like a strange project to me. With just a little effort, you can really make equations that almost perfectly model any curve you find in a photograph.
What's so notable about this? I guess the pictures are pretty.
I agree, I really like the idea but the 2D ones seem a bit forced to me - not that aesthetic. The 3D ones on the other hand are beautiful and have a much more natural feel.
I think it's important that the graphs don't perfectly fit and that the equations look simple (though some aren't really). It might be taking her work too far, but if you think of art as a way of communicating, it's nice to think of messages about how math can guide the real world, but rarely does the real world fully fit.
As an amateur photographer, I have often noticed how the best nature images are often of curves, or other geometrical shapes with some amount of symmetry. Lines that converge to the corners or lead the eye towards the intended focal point are another good example. Seeing function curves overimposed on natural curves tells the observer explicitly of the underlying harmony in nature and in math, which we all perceive, consciously or not.
I don't know about user-end software, but all of that image processing would be fairly simple to do in either Matlab or SciPy. It requires image segmentation (which should be fairly easy in pictures like these with strong edges) and then just running a FFT.
There are some edge-detection algorithms around that are tried and tested. The discrete first and second derivatives are very easy to compute and tend to have very high (or very low) values on edges. Canny Edge Detection is more complicated to implement, but easier to use because it returns the lines themselves, while the derivative operators just amplify the edges, to be picked out with later post-processing (like taking a threshold value). I think there's an online version of the Canny algorithm.
Yes, but you need more than an edge detector. The y-values output by the edge detector might not be a function of x - you could have more than 1 y-value per x, or none at all.
You need a segmentation algorithm to draw a line where the mountains stop and the sky starts, that way y is really a function of x.
http://www.flickr.com/photos/nikkigraziano