Much thanks to Donna for responding.
I'm still getting black areas with Band on a field of scattered points
containing only byte values in the range 0-2 (seems simple enough data set,
right ?). It looks as if some areas that should be assigned the lowest color
value ("0", or green), are pushed off the color scale on the low end to the
background color black. Thinking that it might have to do with the divisions
that I input, I tried varying the "value" list (e.g. adding a limiting lower
boundary as in [-1, 0, 1, 2] or [-.5, .5, 1.5, 2.5], etc., instead of [0,1,2]
which I would expect to be correct), but this did not affect the coloring of
the excluded (black) regions, as hoped. The fact that not all pixels with "0"
value show up as black (only the centers of some singly colored regions), seems
to go against any theory of the specified divisions being skewed somehow. But,
I don't have any other clues.
It did not help to cast the byte data to float prior to band.
I notice that (only) on the problematic byte data, the Band process appears to
increase the number of points in the field ("data items" is a higher number
after Band than prior to). Is this expected behavior ?
You're right, Showpositions does work on the banded data. I'm not sure why I
had a problem with that on my first try, but it's cleared up now.
About the "abutting polygons", does this in fact eliminate interpolation, or
does it just appear that way in the image ? The results to me look like an
"erosion" (standard morphological image processing algorithm). Is it at all
similar ?
On May 21, 10:52am, Donna L Gresh wrote:
> Subject: Re: [opendx-users] Band module
>
> I don't recall you getting an answer to this...
> Perhaps if you can attach a small data file that shows this behavior it
> would help. Also, have you tried casting your data from byte to float or
> vice versa to pin down the source of the problem? you can use Compute to do
> that. I created a very simple program and didn't see any problem per se
> with Band and byte data, so it may depend on something specific to your
> data. Essentially band creates abutting polygons that define the "bands" of
> the quantized data. Each polygon then gets the data value for that band,
> dependent on connections. That's why you no longer see interpolation of
> colors. I'm a little confused by your statement that "showpositions and
> showconnections don't work". They should work just fine on the output of
> Band.
>
> Donna L. Gresh, Ph.D.
> Visual Analysis Group
> IBM T.J. Watson Research Center
> (914) 784-5049
> http://www.research.ibm.com/people/g/donnagresh
> [EMAIL PROTECTED]
>
>
>
>
> Sharon Gibson
> <[EMAIL PROTECTED]> To:
[email protected]
> Sent by: cc:
> [EMAIL PROTECTED] Subject:
[opendx-users] Band module
> son.ibm.com
>
>
> 05/15/2002 01:46 PM
> Please respond to
> opendx-users
>
>
>
>
>
> Is anyone aware of problems or "features" of the Band module ?
>
> I began experimenting with the Band module because I thought that it might
> offer a convenient way to scale float data into discrete byte values for
> the
> purposes of utilizing "delayed" colors. It did work nicely for that, but I
> see
> that it has the additional effect of eliminating interpolation between
> colors
> in the rendering of scattered points. For my purposes, I consider that a
> big
> bonus ! However, I am finding that the results are not perfect in every
> case.
> In particular, when I band byte data (scaling not required, but I'm banding
> it
> so that it looks like the other data sets when rendered), there are regions
> within the image that are black (as if no data). It looks to me like there
> is
> some kind of polygon filling algorithm run amok. The regions that are
> colored
> (not black) look fine (similar to the original, unbanded data), but they
> are
> outlined with the color that would, if correct, correspond to the adjacent,
> black areas. This scenario hints at an algorithm that tries to estimate a
> polygon but closes it inward (forming an outline around an existing "blob"
> rather than outward, to form a new blob.
>
> Strangely, I see no problem with the banded data if it originates as float.
> Or
> maybe the important factor is that the float data has a larger range - the
> byte
> data has values only 0,1,2.
>
> How can I eliminate the black areas ? How does this module work ? I see
> that
> whatever it is doing, it changes the positions/connections somehow (can't
> later
> use ShowPositions !)
>
>
>
>
>-- End of excerpt from Donna L Gresh
