Welcome! We're glad to have you.Originally Posted by Geoff
ImageMagick (a commandline utility) is what I use. It does the best job of any that I've seen. Irfanview is a GUI program that also does very well. Most of the typical programs (Photoshop, Gimp, et al.) do a very poor job in comparison.Originally Posted by Geoff
That said, I should point out that the term "binning" means something different to everyone. The definition you are using is something like "a certain simple method for resizing the image". Other people use "binning" to refer to "resizing the image by any method (specified or not)". It also can refer to on-chip binning (used e.g. on monochrome astrophotography sensors).
Resizing (i.e. resampling) image files is a well-studied area, and there are many other methods that are far superior to the simple 2x2 binning method. All the methods strike a balance between detail and artifacts. The best algorithm is used by ImageMagick above, which is why I recommended it.
It does a full-resolution demosaic, then it downsamples to the smaller size. Unfortunately, the downsampling algorithm is not that great, so the noise level and artifacts are worse than they could be with the ideal software. You can read more about sRAW in my post here on a separate forum:Originally Posted by Geoff
Canon sRAW considered harmful
That is the same conclusion I get from my tests.Originally Posted by Geoff
Keep in mind that downsampling should not be used for noise reduction. It is very bad at it because it throws away noise and detail at the same rate. Of course, if you're printing a certain size, you will have to downsample. But if you don't have to, it is always better not to. The ideal solution is to use good noise reduction algorithms, because they reduce noise faster than they reduce detail. (Using no noise reduction at all is also a valid choice.)Originally Posted by Geoff
Yes, I agree. The only other additional advantage is that sRAW can be processed faster since the demosaic and resample is already completed.Originally Posted by Geoff
Just process the raw files normally and resample with a quality program (like the above) when you're done. The lower quality programs (like what Canon uses in-camera for sRAW) cause aliasing -- which means higher spatial frequencies (i.e. smaller input pixels) with their correspondingly higher noise power contaminates the lower spatial frequencies (large output pixels), which results in more noise than the ideal resampling program.Originally Posted by Geoff
No, because software already has far better algorithms than "true binning". On-chip binning, on the other hand, often has an advantage over any software binning: it may have as much as half the read noise (in the case of 2x2). But it's difficult to pull off with Bayer sensors (artifacts), so aside from one MFDB (Medium Format Digital Back), I'm not aware of any that are doing it yet.Originally Posted by Geoff
Originally Posted by Geoff
The difference in noise between your two comparisons is far greater than it should be. It looks like something in the raw conversion of the full raw file (e.g. noise reduction setting) did not match the sRAW. Can you post the raw files? (Yousendit.com is a free file hosting service - put in any email address and it will give you a link that can be posted here.) If you post them I can do a comparison using other raw conversion software and downsampling algorithms for your comparison.




Reply With Quote