I experienced a similar phenomenon when zooming in on an RGB image. I was zooming
on red roof and at a certain point the colour actually shifted enough that it
didn't look like the same roof. Not sure whether I hit a texture cache limit and
one channel was just no longer available or whether the scaling of the channels
led to it.
We need to keep the averaging for reduced resolution LODs. If anything, if we
could afford an even more sophisticated resampler the best would be a sync
interpolation. If done in the context of APP we might be willing to put up with
the time it takes to generate. Anyway, one decision could be to retain scaling set
at lower LOD levels but this might become less than satisfactory as the absolute
limits change significantly while zooming in. Another suggestion would be to
change the scaling limits calculation so that it uses some lower LOD data and a
sampling of 1:1 LOD tiles. You could still miss the extremes in the data but might
be more likely to get a consistent upper and lower bound.
Frank Warmerdam wrote:
> I wanted to raise an issue we have encountered, though I don't know what
> the solution is at this time.
> The problem seems to be that averaging of noisy radar images produces
> overviews with much lower overall complex value magnitudes than the raw
> imagery. When looking at image data in magnitude mode you tend to get
> a very dark image in overview, often to the point of it being so black the
> image is effectively invisible, but as you zoom into full res, and start
> seeing unaveraged data the contrast becomes much better.
> We see a similar effect when looking at just real or imaginary values, though
> the averaged values (near 0+0i) produce a nearly uniform grey.
> Most disconcerting of all is when you zoom in on these images. As the
> system changes the overview level being used, the average intensity of
> blocks changes substantially. In overview mode individual blocks that
> are still being displayed from full (or higher res) textures show up
> as alarmingly bright.
> So, my question is what should we do about this? Using nearest neighbour
> resampling instead of averaging the problem would go away, but there are
> very legimate reasons to use averaging with radar data overviews.
> Best regards,
> I set the clouds in motion - turn up | Frank Warmerdam, warmerda@...
> light and sound - activate the windows | http://members.home.com/warmerda
> and watch the world go round - Rush | Geospatial Programmer for Rent
> Openev-discuss mailing list