Outside shot... ( First Exterior Red frame grab )

Graeme_Nattress said:
TS1 - wish I could be what we're doing is proprietary, and also designed purely to be used with our sensor. I couldn't call it a generic algorithm. I did compare what we're doing with Adobe RAW converter and thought ours looked slightly better on our images....

Graeme
Click to expand...

Graeme,

I understand your constraints. Perhaps we can limit our discussions to public domain algorithms, and you could advise us on selecting/tweaking published schemes.

Does RED also do some sort of joint demosaicing and denoising for use with higher ISO images? A bunch of such algorithms have been recently published and promise better results than performing the two steps independently.

And finally, what is special about Mysterium that requires a specialized/tweaked algorithm? Is it the anti-alias filter/CFA pattern? Or is that confidential too?

Cordially,
Tripurari
 
Happy to talk public stuff and demosaic theory with you!

We're not doing any denoising in camera. I think that's best left to post where you can carefully control it. I did have some demosaic algorithms that performed better or worse with regard to noise performance, so yes, doing those both together can be advantageous!

There's nothing especially special, but I did tweak some "magic numbers" that are sensor and OLPF dependent.

Graeme
 
mycc.jpg
 
Graeme,

Thanks for your generous offer, I'll seek your advise when I need it.

About not denoising in the camera, how well does REDCODE handle noisy RAW data? The big problem with demosaicing and denoising separately is that the demosaicing step substitutes the random nature of the noise with artificial patterns. These patterns can't be filtered out by the denoiser. This *could* be a potential problem with the REDCODE raw compressor as well.

http://www.accidentalmark.com/research/index.html has a bunch of papers on the topic of joint demosaicing and denoising.

K. Hirakawa, X.-L. Meng, "An Empirical Bayes EM-Wavelet Unification for Simultaneous Denoising, Interpolation, and/or Demosaicing," IEEE ICIP, 2006.

appears to yield the best results.


Regards,
Tripurari
 
Well, it's hard to get noisey data off the mysterium to test this :-) but seriously, the algorithm is designed to cope quite well! REDCODE RAW has it's own methods of dealing with noise :-)

Graeme
 
Graeme_Nattress said:
Well, it's hard to get noisey data off the mysterium to test this :-) but seriously, the algorithm is designed to cope quite well! REDCODE RAW has it's own methods of dealing with noise :-)

Graeme
Click to expand...

Ha! I wish I could say the same about my still camera.

Jim once wrote that it would be a while before such a clean sensor makes its way into a competing camera. Do you guys have patents to exclude other manufacturers? If so, I'd like to read them.

Regards,
Tripurari
 
you should liscense that chip to Nikon! They need it in the Dx3!! though the Dx3 will need to be around 18-24 Megapixels to warrant a new model... but still.... or is a DSLR a secret development project at Oakley as well? if thats the case then make sure to use the F-mount... that way i can get one!

:~)
 
You must log in or register to reply here.
Back
Top