Re: Is camera response needed for HDR based upon multiple exposures ?
- From: aruzinsky@xxxxxxxxxxxxxxxxxxxx
- Date: Fri, 12 Oct 2007 13:11:00 -0700
On Oct 11, 6:28 am, haim <haimg...@xxxxxxxxx> wrote:
Hello,
In many papers it is suggested to recover first the non linear
responese f(E) of the camera prior to the blending of images with
different exposures. Given f, the inverse of f is estimated and the
mixing of the images is carried out in that domain. On the other
hand, many procedures to improve the dynamic range are based on
simple rules working on the original images without any estimtion of
f. Could somebody clarify that point ??
Haim
On Oct 11, 6:28 am, haim <haimg...@xxxxxxxxx> wrote:
Hello,
In many papers it is suggested to recover first the non linear
responese f(E) of the camera prior to the blending of images with
different exposures. Given f, the inverse of f is estimated and the
mixing of the images is carried out in that domain. On the other
hand, many procedures to improve the dynamic range are based on
simple rules working on the original images without any estimtion of
f. Could somebody clarify that point ??
Haim
No, can you provide links to those papers?
Linearity is practically only important for postprocessing. Some tone
mapping methods are visually insensitive to non linearity therefore
linearity can be neglected in those cases. Optimizing signal to
noise ratio is the important part and that is whatsome authors
apparently neglect.
I have recent experience (reinvented wheel) with that and my methods
will be incorporated into the next release of SAR Image Processor.
Stacking frames with different time exposures, by simple averaging,
maximizes signal to noise ratio under the assumption of an additive
white noise model with constant variance, and it is not necessary to
know the variance or f(E). Also, note that the signal varies beteen
pixels and this is also unimportant. As far as I know, the only
significant flaw of this model is that it does not take quantization
and multiplicative noise into account. However, I can tell you that
including quantization noise in the model would require knowing
variance of the additive white noise to optimize. A simply stacked
result will typically have a very non linear response (much worse than
f(E)) that looks very bad without further processing. My procedure
involves adjusting the histogram of the simply stacked result to match
that of a more linear, but, noisier image. You should be able to
invent your own ways of doing this.
.
- Follow-Ups:
- References:
- Prev by Date: Re: child molester's disguised face unwound
- Next by Date: Re: Is camera response needed for HDR based upon multiple exposures ?
- Previous by thread: Is camera response needed for HDR based upon multiple exposures ?
- Next by thread: Re: Is camera response needed for HDR based upon multiple exposures ?
- Index(es):
Relevant Pages
|
