documenting ops. durand

[Alexandre Prokoudine]

Hi,

Recently I started working on a new user manual for Qtpfsgui. I have
to admit that I'm neither programmer nor a math-head person, so I'm
doing my best, but still missing some clues :) So I'd really
appreciate some hints.

The first op I have a question about is Durand. Range kernel sigma
isn't really mentioned in the paper explicitly (maybe it's in
equations, but I'm lacking skills to read them). So, the only thing I
figure out is that larger values produce slightly darker images with
less homogeneous distribution of pixels across the tonal range and
much increased halo around edges. How could I best describe this
setting?

Alexandre

[rafm]

Hi Alexandre,

Durand et al. TMO is based on the bilateral filter, which splits an
image into two layers: one with smooth gradients and sharp edges (base
layer) and the other with textures and fine details (detail layer).
Then the contrast of the base layer is compressed. The range kernel
sigma decides what is the mean amplitude above which the edges are
preserved in the base layer. If you set this parameter too high, only
very high contrast edges will be preserved in the base layer, while
the other edges will be smoothed out, creating halos in the resulting
image. If you set this parameter too low, everything will end up in
the base layer and the resulting image will loose sharpness (because
base layer gets compressed).

The ideal behind this TMO is to compress large edges while preserving
small edges and the range kernel sigma gives a fuzzy threshold which
edges should be considered as large and which as small.

If I were to design a GUI, I would either hide this parameter (default
value works well for most images), or call it "preserve edges lower
than" and put it on the log scale from "small" to "large".

Cheers,

Rafal