# How to count EFFECTIVE number of pixels in a photo

Started 5 months ago | Questions thread
Re: How to count EFFECTIVE number of pixels in a photo

Thank you for the lovely debate, you really help me to get a better understanding of the matter and I am learning as we speak. 🤦🏻‍♂️😂

ProfHankD wrote:

MTFMapperon an ISO sequence of test images. In theory, loss is the same independent of scene photographed, although in practice that's not quite true because MTFMapper computes based on a monochrome image. Resolution is very color dependent for cameras using a CFA, nearly 2X for green what it is for blue or red; for that matter, lens resolution generally decreases for longer wavelengths due to diffraction, so red resolution is typically less than blue. If you're trying to be very precise, this can be a very deep rabbit hole.

Yep, I am not convinced that this artificial metrics with b&w rectangular patters can be representative to what happens in real low light conditions with artificial light.

This is what it said:

Here's a step-by-step approach:

1. Calculate the entropy of the original image.
2. Calculate the entropy of the denoised image.
3. Compare the two entropies to determine the percentage of detail retained.
4. Use this percentage to determine the effective resolution.

Can't get my brain around this, I can think about corner cases that will brake this logic, but maybe I'm wrong and should go for it. 😂

I might try asking it to write a python or Octave scripts based on this idea for me.

As usual for ChatGPT, that sounds reasonable and is complete nonsense.

The ChatGPT suggestion is both difficult to compute and pointlessly wrong.

I need to rethink this whole approach over. Maybe not entropy, hope I will come up with a more practical approach than the artificial b&w mtf metrics that's a real pain to measure.

-- hide signature --

Cheers,
Alex

il_alexk's gear list:il_alexk's gear list
Pentax K-5 IIs Sony a7R III Fujifilm X-A5 Fujifilm X-T100 Pentax smc FA 50mm F1.4 +9 more
Complain
Post ()
Keyboard shortcuts: