Great suggestion, Curtis. But my recent strategy has been to
keep the core VisAD functionality stable, so I think it would
be best to do this in the application. Hope that's not a
On Wed, 18 Feb 2004, Curtis Rueden wrote:
> As long as I'm sending lengthy ramblings to the list, I thought I'd
> bring up another issue as well.
> VisAD's autoscaling support works very simply, by mapping the lowest
> sample value to min, and the highest sample value to max. This is the
> most "honest" way to do autoscaling. But other ways may sometimes
> produce more desirable results, depending on circumstances.
> For example, I have a FITS file that has one tiny spot with values
> somewhere in the 40,000 range, while the rest of the samples range from
> 0 to 100ish. The current autoscaling produces an image that is
> completely black, with one tiny white spot. Cases like this occur often
> with microscopy images.
> To produce a better initial image in the above case, one could compute
> the sampling mean and standard deviation, then use those values to
> create an intelligent color mapping that emphasizes sampling differences
> without being overly influenced by outliers.
> Even in the case of spatial mappings, it may occasionally make sense to
> do things this way. With the above FITS file, mapping the image range
> RealType to Display.ZAxis would produce a nice surface, except for the
> spike where the outliers are. Instead, the resultant display is
> essentially flat with a single spike. If the autoscaling used some kind
> of statistical approach, we'd see the relevant surface instead.
> My idea is to expand the autoscaling logic to account for more than one
> autoscaling strategy. For example, there could be a method of ScalarMap
> and/or GraphicsModeControl such as:
> public void setScalingStrategy(int strategy)
> where strategy can be MINMAX_STRATEGY (the default, as it is now),
> STDDEV_STRATEGY, or possibly other alternative scaling techniques.
> To implement this, the loop that computes min and max values would need
> to also compute some basic statistical information such as mean,
> standard deviation, and possibly other values such as median and mode.
> But such changes wouldn't affect efficiency.
> Of course, I could code this statistical color scaling logic in VisBio,
> and just use ScalarMap.setRange() to implement it. But the functionality
> seems generally useful enough that it might make sense for it to be part
> of core VisAD.