Re: Transforming a sparse histogram to a Gaussian histogram

On May 3, 1:56 pm, ImageAnalyst <imageanal...@xxxxxxxxxxxxxx> wrote:
On Apr 30, 10:22 pm, Dev <gun...@xxxxxxxxx> wrote:> Hi All,

I want to convert a natural image (which has a sparse historam
distribution) to an image with a Gaussian histogram distribution. I
tried the gamma correction, stretching and other simple methods but
they didn't work. Could anyone please provide me an algorithm\matlab
code to do this?

Thanks,

Dev

---------------------------------------------------------------------------­----------------
Dev:
I don't think it can be done satisfactorily in general.  Yes,
mathematically it's possible by pulling a few tricks but it won't look
like what you want.  For example, let's say your sparse histogram has
gray level bins populated at 50, 150, and 250 (that's pretty sparse
but it will illustrate a point).  Maybe it's just a gray level step
wedge - an image of three gray rectangles.  Now, let's say that you
want a Gaussian of mean 200 and standard deviation 5.  And let's say
you came up with a method to produce that image.  That will look like
a uniform bright image (mean of 200) with a slight noise (the stdev of
5).  Well that looks nothing like what you started with, so what good
is it?  Yeah, maybe you can finagle it so that the one step is
comprised by all the gray levels 195-200 and the other step has gray
levels mostly in the 200-205 range, but again, what use is this?  Why
would that be an improvement over your original image?

Please give me the background of what your REALLY want to do.  You're
thinking that getting a Gaussian shaped histogram will do something
for you.  What is that?  Why do you think this will accomplish what
you REALLY want to do?

Perhaps what you're really after is color temperature conversion or
histogram matching.  An excellent paper on histogram matching is here:http://www.cl.cam.ac.uk/~mg290/Portfolio/ColorHistogramWarp.html

Perhaps you have an image with a "posterized" appearance to it - large
regions of uniform gray levels with sharp edges between the uniform
regions.  Do you want to smooth out those boundaries to make them less
noticeable?  This is what I mean when I ask you what you really want
to do.  If you tell me what you really want to do then I can suggest a
remedy - and it might not necessarily be the Gaussian histogram
solution you picked.

Looking forward to your answers....
--ImageAnalyst

Hi All,

Thanks for all the replies. I am trying an algorithm on image
restoration which uses a Gaussian distribution. As natural images have
a sparse distribution rather than a gaussian distribution, I have to
convert the image to have a gaussian distribution which matches the
sparse distribution. I will go through the provided links in more
detail. Meantime if you have any other examples/code/documents, please
let me know.

Thanks,

Dev
.

Relevant Pages

  • Re: Transforming a sparse histogram to a Gaussian histogram
    ... distribution) to an image with a Gaussian histogram distribution. ... Usually the implicit assumption for image restoration is additive Gaussian *noise*. ...
    (sci.image.processing)
  • Re: Transforming a sparse histogram to a Gaussian histogram
    ... distribution) to an image with a Gaussian histogram distribution. ... Usually the implicit assumption for image restoration is additive Gaussian *noise*. ...
    (sci.image.processing)
  • Re: Transforming a sparse histogram to a Gaussian histogram
    ... restoration which uses a Gaussian distribution. ... you're trying an algorithm on image restoration. ... the degraded images have a Gaussian histogram. ...
    (sci.image.processing)
  • Re: marching cubes help
    ... and concise question to which you expect a clear and concise ... I mean that the distribution is radially symmetric about its ... of Gaussian blurring of a 3D voxel image. ... leading to a large number of small connected components of triangles. ...
    (comp.graphics.algorithms)
  • Re: Hashtable performance metric
    ... this histograms shows a uniform distribution. ... Your chi^2 value chisq is computed by ) ... your hash function. ... whether the observed distribution (the length histogram) is within the ...
    (comp.programming)
Loading