[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Sorting and image rescaling

Bill <wclodius@lanl.gov> writes:

> I work on some of the software for an imaging sensor. One of the codes I
> wrote was to generate JPEGS. In order to generate prettier JPEGS I wrote
> my own bytescaling routine that crudely takes into account the
> distribution of values within a band image. This routine relies on a
> single line equivalent to
>      sorted_values = SORT(band_image)
> where SORT is the IDL intrinsic, and band image is a floating point two
> dimensional array.  I basically use this line to find a set percentage
> of minum and maximum outliers , and use the maximum and minimum of the
> remaining inliers to do a linear rescaling to values from 0 to 255.
> Some of our band images are on the order of 2500 by 10000. For such band
> images this line can take over 30 seconds per band. This is a moderate
> nuisance at the moment, but we are planning to update our calibration ,
> and reprocess 1000s of multiband images with a new calibration.
> Naturally we want to update the jpegs to reflect this new calibration.
> It appears that this single line will extend reprocessing by a couple of
> days. I don't like this. This yields the following questions:
> 1. Does anyone know a better general approach to such a rescaling that
> avoids the need to sort the data, or sort more than a fraction of the
> data?
> 2. How does ENVI do its linear, gaussian, and uniform rescalings? They
> seem to take about a second for these images, so they must be doing
> something different from what I am doing.
> 3. Does IDL have a particularly inefficient SORT method for floats? Note
> that for floats it is possible to sort in 0(N), using something like a
> bucketsort, but more flexible sorting routines such as merge sort, heap
> sort, and quick sort are of order 0(N ln(N)).

This sounds like a prime time application for histogram ;-) Just don't ask me how to use this magical hat ;-(


[[ Dr. Martin Schultz   Max-Planck-Institut fuer Meteorologie    [[
[[                      Bundesstr. 55, 20146 Hamburg             [[
[[                      phone: +49 40 41173-308                  [[
[[                      fax:   +49 40 41173-298                  [[
[[ martin.schultz@dkrz.de                                        [[