Sunday 12 January 2014

Pixels to Perceptual Mega Pixel (data review)

I thought that I'd follow through with some of the data I've compiled to me suggests that to increase the effective (or Perceptual) megapixels on needs increasingly more actual pixels with each sensor format but that with increase the return value is small.  So with a smaller sensor like a APS or 4/3 sized sensor you need to have way higher increases of pixels to get perceptually much increase.

camera MPix/P-Mpix Native Mpix
GH1 (4/3) 1.33 12
GH3 (4/3) 1.45 16
40D (APS) 1.68 10
7D (APS) 1.99 18
5D (FF) 1.16 13
1DsIII (FF) 1.4 21
5D III (FF) 1.3 22
Sony RX1 (FF) 1.33 24

With micro4/3 some quick observations (which seem to be representative) are found in such lenses as the Olympus 75mm f1.8 and their 45mm f1.8. These lenses give 9 and 7P-Mpix (respectively) on the GH1 and 9 and 11P-Mpix (respectively) on the GH2. For the This is a ratio of 1.33 and 1.44 Pixels/P-Mpix for the Oly 75mm lenses on the GH1 (12Mpix) and GH2 (16Mpix) bodies, demonstrating that as pixels go up the 'return ratio' of perceptual megapixels gets less. This is otherwise known as diminishing returns.

I suspect that the data on DxO is insufficient (unverified and uncross referenced tests) for a good analysis (to allow for removal of measurement error by proper statistical samples). From my reading of the data on the APS cameras the results seem to follow the pattern. I expect diminishing returns will flatline when you ultimately hit the "aerial resolution limits" of the lens (as one would normally do in MTF testing)

With the Full Frame stuff the 5D was the low hanging fruit, with a very good ratio of yeild per sensor pixel. However as the desire to increase output quality went higher the requirement for capture pixel to Perceptual pixel ratio when higher too. The Sony RX1 for instance has 24Mpix to get its 18P-Mpix

It would seem to me that an appropriate 'storage' compression here would be to downscale the captured RAW file into a DNG (or the like) from the captured Mpix to the reasonable limit of real IQ. There would be little or no loss as observed in my previous post.

Back in the past designers seemed to not be restricted to sensor size and just moved up to a slightly larger film format 35mm - 645 - 6x7 - 6x9 ... and larger

Perhaps the best answers here are really the same?

I would love to see some tests of the Leica S2 system.

2 comments:

jann0k said...

Thanks to your previous article I now understand dxomark P-MPix rating.

You are right that it doesn't make sense to add more pixels into m43 system. And it also means that FF or larger is better when big prints are needed.

But dxomark rating system is not good for comparing lenses between different systems as at first it seems that FF lenses are the best (in resolution and dxoscore). But because m43 sensor is 4x smaller (in size) than FF sensor, then 16MB m43 sensor resolution is actually equal to 64MB FF sensor resolution.

Thats why 5D gets very good ratio as even medicore lenses outresolv its sensor resolution. To compare lens resolution (lpmm) on dxomark they need to be mounted on similar resolution sensor eg 16MB m43 and 24MB DX sensor and then adjust MPix ratio to get comparable lpmm. But dxomark PMpix only shows center sharpness at sharpest aperture. I like lenstip review more for comparing lenses on different systems as they display lpmm on different apertures on center and edge.

Of cause for system (lens+sensor) dxomark PMpix is about correct as higger Mpix pictures can be printed larger. At a same time for web 12MB sensor is more than enough as 4K (~4MB) displays are now appearing and most of people are using 2K displays. I also have 100cm x 70cm photo printed from 12MB Nikon 5000 sensor and from normal viewing distance its more than good enough.

obakesan said...

Hi Jann0K and thanks for your comment.

Actually I was going to address the point you raised "But dxomark rating system is not good for comparing lenses between different systems as at first it seems that FF lenses are the best (in resolution and dxoscore). But because m43 sensor is 4x smaller (in size) "

which is why I included the 35mm f2 lens with the 7D where it only scores 9P-MPix (unlike the score of 11 that it gets on the 5D and 17P-Mpix on the 5DmkII). Obviously as long as the body and lens combo is present you can get a clear idea, however you can still sort of sus out how it will perform as the score of 9 on the 7D is a similar ratio to cropping APS out of the guts of a 5DMkII.

Obviously MTF tells a better story, but only for those who can read that language :-)