Re: Manipulate Histogram to get specified mean, stddev
From: Piotr Stanczyk (piotr-s_at_NOSPAM.moving-picture.com)
Date: 11/29/04
- Next message: Han de Bruijn: "Re: What can we expect by taking Fourier Transform of noise or random samples?"
- Previous message: kiki: "how to compensate for low dynamic range monitors?"
- In reply to: Martin Leese: "Re: Manipulate Histogram to get specified mean, stddev"
- Messages sorted by: [ date ] [ thread ]
Date: Mon, 29 Nov 2004 10:05:43 GMT
Having re-read the original post, I rather think you may be right on
that. Thanks for point that out.
Piotr
Martin Leese wrote:
> Piotr Stanczyk wrote:
>
>> You will find quite a good treatment of this in the book by Gonzalez
>> and Woods, Digital Image Processing.
>>
>> I'm going by memory on this, but this can be viewed as a histogram
>> matching problem. So, this is usually explained in 2 steps.
>>
>>
>> Let T(x) be the cumulative probablity distribution function of the
>> input image. Get this by summing up the pdf, ~historgram, to x
>>
>> -- x
>> \ hist(x) = T(x)
>> /
>> -- 0
>>
>> Applying just T(x) gets you just the histogram equalisation process,
>> from here you can apply the mapping that will give you the
>> charactersitics you require ...
>>
>> Let G(y) be the cumulative probablity distribution function of the
>> output image. Since you have the pdf given to you via the mean and
>> stddev, you can construct a Gaussian pdf with these values
>>
>> Then the tranformation you want is:
>>
>> inv{G} (T)
>
>
> This will not only change the mean and stddev, but
> will also force the shape of the histogram to be
> Gaussian. My guess is that this is not what the
> original poster wanted.
>
> Jon Campbell's post contains a formula which
> changes the mean and stddev, and also preserves
> the shape of the histogram. Please ignore the
> formula in my previous post to this thread as it
> is wrong.
>
> The original poster asked for a look-up table. To
> create this, simply apply Jon's formula to values
> from 0 to 255 and store the result.
>
- Next message: Han de Bruijn: "Re: What can we expect by taking Fourier Transform of noise or random samples?"
- Previous message: kiki: "how to compensate for low dynamic range monitors?"
- In reply to: Martin Leese: "Re: Manipulate Histogram to get specified mean, stddev"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
