One of the features of the
latest generation of film scanners is the ability to scan the
same image multiple times and average the results. This is
commonly called multisampling. The goal is to reduce noise in the
darkest parts of the original. This corresponds to shadows in
transparencies and highlights in negatives. Noise in the darkest
shadows is probably less objectionable since the eye is less
sensitive to small variations in brightness. In highlights,
however the noise may produce a grainy effect.
In an attempt to answer the question as to whether multisampling produces any noticeable effect I have run this experiment. The image at right is a scan of an empty slide mount done using a Minolta 5400 scanner set at maximum resolution and with no image adjustments. We would expect the top to be black, but correcting color balance is a type of image adjustment. The important thing is that neither the dark or light parts reach the limits of 255 or 0 so all data is preserved. The data was assigned the correct color profile as I would do with any image before editing.
|Measuring around in the light area
the readings are RGB = 251,252,253
without any variation.
In the dark area there is this variation: R=68 or 69, G=21 B=17 or 18.
Using 16 bit values gives R=6970 to 6992, G=465 to 467 and B=1421 to 1451.
This amounts to a 0.3% variation for the Red and a 2% variation for the Blue.
A histogram of a portion of the dark area is shown at right taken from the original 16 bit scan. Note that Photoshop reads in 8 bit values even for a 16 bit image.
After converting to 8 bits the histogram looks the same. So it would seem that the loss of information will not change the visible appearance.
The thing to note is that whatever noise is present is going to be displayed as a change of value of 1 or 2 at most, and especially for dark areas will have minimal perceptual effect.
Dark Area Histogram
image at right is corrected so that the dark areas are all about
10 or 11 and the bright about 252. These are the values I would
usually use when printing on an ink jet printer. Setting the dark
to 0 doesn't produce any more density on the print but makes the
other shadow tones too low in contrast and dark.
Since actual film is never going to be this opaque or transparent I would have to conclude that scans of film will also be essentially noise free and thus multisampling in this case will not yield much of visible value whether the data is captured in 16 bit mode or not. I still suggest 16 bit scans so that there is less worry about posterization when doing large-scale tonality adjustments.
If anyone has an example to the contrary I'll be pleased to reference it here or add to this discussion.
Note the sample are shown at full size with no compression so you can download them and measure them yourselves if you wish.
© 2004 Robert D Feinman