>>>But maybe before going into details, can you think of another way of spacing out points along a curve so the the distance between each pair of points is roughly proportional to the local luminance of an underlying image? Maybe there's a straighter path that I'm not thinking of.
If you have a mathematical formula for the "raster gradient" of the luminance, that could perhaps be used to array the points along the curve.
The programs that create the raster gradients use some mathematical method. See their source code, or do internet search?