Canon 600/4 for $12k.....

Started Feb 7, 2011 | Discussions thread
Joofa Senior Member • Posts: 2,759

Hi Steen,

The problem is that the gang does not seem to appreciate that unless the methodology for specifying how the pixel count normalization is specified then it is not very helpful to talk about the "light falling on the sensor" mantra. That mantra, being theoretical, will only take one so far. In practise we have image samples resulting from sampling that light and any operation such as display normalization, snr calculation, etc. must come from those samples.

I asked Great Bustard that how does noise compare if I do pixel count normalization by:

(1) keeping top left pixel in a block,
(2) keeping bottom right pixel in a block,
(3) average,
(4) Lanczos.

And, more importantly how do we get a quantitative analysis of noise vis-a-vis signal in such operations. The reason I specified the above 4 methodologies is that it is a tenant of equivalence, as I understand, that one should normalize for same size if the pixel counts are different. However, I never got an answer from him, except, handing it over to Bob.

Now, consider the image below.

On the left is an image that would have been seen by a certain sensor. Now suppose I want to compare this with an image obtained from a sensor with 2x2 smaller spatial extent, but same pixel size, resulting in different pixel counts on both sensors. Before, I go and acquire an image with that smaller sensor, there is a very important consideration that the gang is missing. On the top right is an image that I obtained by methodology (1) and on the bottom right is an image that I obtained by methodology (3). Now, before even I have a chance to compare to an image acquired from the smaller sensor, I have an issue that the 2 methodologies that I have chosen for pixel count normalization are giving me two very different images, with very different noise properties. Which one to pick? One thing is clear, that different methodologies that I chose for pixel count normalization for comparing between bigger and smaller sensors does have an impact.

This is one of the core issues that I see that the gang fails to appreciate regarding the notion of pixel count normalization in generating same sized views when they talk about equivalence.

And, we haven't even started talking about any quantitative analysis of the properties of noise in such operations yet.



-- hide signature --

Dj Joofa

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