I've uploaded a picture cleaned using my method.

You can find the original slide as scanned at
http://ljl.150m.com/slide_original.jpg

The cleaned picture is at
http://ljl.150m.com/slide_clean.jpg

The "noise map" as scanned (lid open, little ambient light) is at
http://ljl.150m.com/slide_map.png

There is post-processed version of the noise map, which is the one
actually used as the alpha channel, at
http://ljl.150m.com/slide_nmap.png


All the scans were made at 1200 dpi, 24-bit color with an Epson RX500.
The "image scan" and the "noise scan" have been hand-aligned, since my
scanner doesn't pick up the same image area in film mode as in flatbed
mode.
The final images have been scaled down (Paint Shop Pro, "Pixel
Resize") to be 800 pixels wide.

As you can see, I didn't get the histogram stretching quite right for
the "normalized noise map": I made the darkest 97% of pixels black,
and the brightest 0.1% white, which seems to be definitely overkill. I
need to experiment a bit more with this.

You can see how I also applied a convolution (whatever a convolution
is) to dilate the noise spots, in order to avoid the edges of the dust
particles to show up in the final image.


by LjL