From DXOMark: More pixels offsets noise3

Started May 30, 2009 | Discussions thread
natureman Veteran Member • Posts: 3,979
Re: Terms - not too confusing

Bogdan_M wrote:

Steen Bay wrote:

The terms and language used in these discussions can sometimes be
pretty confusing, at least to me. For example I managed for a long
time to misunderstand what you and others meant by the term 'image
level'. When I was told that at 'image level' the noise was largely
the same, regardless of the pixel count (same sensor-size), then I
took it to mean that the total noise would be the same, when
comparing images at the same output size (not 100% view), like in big
prints.

Later Bob (bobn2, who unfortunately has been permanently banned
now..) told me that comparing at 'image level' was comparing images
at the same level of detail, meaning that the image with more MP's
would have to be downsized/blurred (and thereby lose most of its
extra detail) for the noise to be the same, again meaning that there
wasn't really any 'free lunch', as i thought at first.

These two things are the same. Image level for him was normalisation
(by downsampling with preblur) and then comparing the normalized
images at 100% (because once normalized they have the same
magnification).

How can you consider that the same magnification? If one image is downsampled (or upsampled), and the other is not, to match the sizes of the images, the magnifications are not the same.

Let's say there are two rooms. One is ten feet by six feet, the other is ten feet by ten feet. If you're looking at the entirety of each room (100%), are they the same size? If you hide or remove a portion of the ten by ten room so that it matches the ten by six room in size, are you still seeing 100% of the ten by ten room?

You could compare them at any size, but at 100% (once
normalized) you have the finest level of detail.

Now, the problem with downsampling is that you can do it in 2 ways,
by preblurring the image (which will lose a bit of resolution but
will also reduce noise) or witout, case in which images will look
(about) as noisy, but will have more detail. The "free lunch" par
meant just that you can't visibly do both at the same time, enhance
detail and decrease noise, at least not unless there's a very big gap
in the pixel count between the two cameras

By the way, sorry to find out he was banned...
--
http://www.flickr.com/photos/bogdanmoisuc/

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