Which of these two metrics is a good measurement of sharpness?

Question

I have seen two metrics used to measure the sharpness of a camera:

• Perceptual megapixels, used on Dxomark.
• Number of lines the camera can resolve in a resolution chart.

However, these two sometimes give completely different results. For example, the Lumix LF1 has a perceptual megapixel count of 3MP, while the Sony RX1 boasts 18MP -- a 6x difference.

But when we compare the number of lines the camera can resolve on a resolution chart, a completely different picture emerges: The Lumix does 22, while the RX1 does only 26. This is only an 18% difference, which is worlds away from the 6x difference we calculated earlier.

I can understand a slight difference between the two measurements, but not 18% vs 6x. Where does this difference come from? Horizontal vs vertical resolution? Sharpness at center vs at edges?

Which number is more useful as an objective indicator of sharpness? If I take a photo using both cameras and view them on a 4K monitor, after downscaling both photos to the monitor's resolution, will I see a huge difference? Going by the perceptual megapixel number, since the Lumix manages only 3MP, while the 4K monitor is 8-9MP, and the Sony does 18MP, yes, I should see a world of difference. But if I go by the resolution chart, the conclusion is that there will only be a small difference.

Notes:

1. I'm comparing both cameras at their base ISO, and in either RAW or JPEG -- whichever is sharper.

2. I know that sharpness varies between the center and the edges, but let's ignore this detail for now. If a camera has a certain perceptual megapixel count, I don't care how those are spread across the sensor. I just care about the total number of perceptual megapixels.

3. I'm comparing fixed-lens cameras in this example to avoid confusion about which lens is being used.

4. I'm ignoring Snapsort's true resolution, since it assumes a perfect lens. I'm interested in real-world measurements of the camera's sharpness, which should take the lens into account.

5. I'm also ignoring MTF, since it seems like the perceptual megapixel number is just a summary of it.

Answer 1

I'll just add, the two numbers measures different aspects.

Dxomark's "perceptual megapixel" are measuring both resolution and contrast, hence the use of the word "perceptual". It's not the actually detail sharpness but how we as humans sees it on print in the size they uses for the measurements.

Dxomark is basically telling you how most of us perceives the results IRL, while MTF shows actually resolution of a lens (very useful for knowing the print size or the cropping capability).

In short Dxomark gives us a way to see how satisfied we would be with the results of the combo, MTF shows the professional usability of the same.

Perceptual megapixels can be raised slightly by post processing, MTF however can't.

Answer 2

However, these two sometimes give completely different results. For example, the Lumix LF1 has a perceptual megapixel count of 3MP, while the Sony RX1 boasts 18MP -- a 6x difference.

But when we compare the number of lines the camera can resolve on a resolution chart, a completely different picture emerges: The Lumix does 22, while the RX1 does only 26. This is only an 18% difference, which is worlds away from the 6x difference we calculated earlier.

Short answer: do not compare MTF50 lines per mm, compare MTF50 lines per frame height. This will give you difference similar to difference in perceptual megapixels. If you square the ratio of "lwph" between cameras you will roughly get the ratio between perceptual megapixels.

Perceptual megapixels are not bad but not too good either because characterising optical resolution with one number is oversimplistic.

Answer 3

Neither. MTF50 (The spatial frequency at which MTF drops to 50%) is.

find a source of test datathat uses the program Imatest for its measurements, because that uses MTF directly.

As for the measures you mentioned, to properly compare "perceptual megapixels" wrt sharpness/resolustion you need to take the square root because megapixels ~ pixels^2.

So the comparison is 1.7 vs 4.2, or 2.5:1. Dxo really does people a disservice but not mentioning that.

I'm not even sure what a "perceptual megapixel" is- the Dxo describes it, it seems like a measure of information loss.

The problem with using number of lines resolvable is that it doesn't measure sharpness, but measures resolution, which is related to sharpness but not the same. Resolution is the spatial frequency at which the MTF drops below a certain level, 10% IIRC.

You can have lens A which is sharper than lens B, but B has better resolution.