Are we still comparing ISO performace at 100%?

Started Jan 25, 2010 | Discussions thread
Shop cameras & lenses ▾
Regular MemberPosts: 267
Re: O.k, I get what you guys are saying, but...
In reply to Slideshow Bob, Jan 31, 2010

Slideshow Bob wrote:

... the problem I have with upsampling is that you usually loose sharpness. First, you don't want to sharpen the 100% image during initial raw processing, because any normally acceptable sharpening artifacts will be enlarged with the upsampling and look pretty ugly. After the upsampling, you can't easily sharpen because the interpolation usually screws up edges, and applying sharpening only exacerbates that problem. Of course, I'm not saying it's impossible to make a decent enlargement (and I do agree with what you say to a certain degree), but increasing the resolution only to loose the sharpness is a somewhat dubious tradeoff IMHO.

What you're saying is the 100% viewer's perspective. If you're upsampling properly, you won't lose sharpness, but you won't gain it either. So, a 12MP camera upsampled to 18MP won't give as much detail as an 18MP camera native. Your comments on sharpening are apposite too. The core problem with the lower res file is it doesn't have enough information in it in the first place, and it suffers from an AA filter which rolls off at spatial frequencies visible in the image (and sharpening is usually done to counteract the AA filter rolloff), so you're always between a rock and a hard place. If you sharpen native, you won't get well defined edges at the output pixel count, if you sharpen after resampling you'll sharpen artifacts. Best, in the ned to have a camera with sufficient pixel count that you're generally downsampling.

Perhaps I would find a tool like Genuine Fractals a more workable solution, but for now I'm very happy with A3 prints from 100% files that have been properly processed.

Which is saying you're happy with the way your printer drivers do the resampling for you. Fair enough. I found I could do a better job than my printer could.

In the context of the OP's post, the losses of up/downsampling are the simple reason as to why it's best to appraise the capabilities of sensors at their native 100% resolution, particularly on an internet forum where we invite others to make comments. At least then an objective inspection on the result can be made, whereas if an image is fundamentally processed to suit one persons needs or taste, it all becomes entirely subjective.

It is subjective in any case. In a viewed comparison you are making a subjective assessment of the relative noise, and relying on a subjective extrapolation of each different camera to any real use case. The only 'objective' is properly conducted measurements of the noise and extrapolation using the well understood theory to the different use cases, as DxOmark does. The fact that DxO is so often decried by the '100%' proponents suggest that the subjective assessment they're making on the first count is at variance with the objective, and almost invariably they fail to make an extrapolation to sensible use cases which is in accord with reality. In the end, 100% comparisons are of little use to anyone in helping to inform equipment decisions. IMO, generally, they lead to profoundly misleading judgements. Their saving grace is that they bolster the self esteem of some poor souls, by helping them convince themselves that the camera manufacturers have made no progress since they bought their camera.

Anyway, iscogottingen, ejmartin, and RDKirk, thanks for taking the time to explain your methods.

You're welcome.

Reply   Reply with quote   Complain
Post (hide subjects)Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow