Does camera sensor size matter?

Introduction

For the longest time, I’ve been wanting to conduct a quantitative survey of whether sensor size really matters. Every photographer I know cares deeply about sensor size, with the general theme being “bigger is better”. Since I have three different sensor sizes at my disposal, along with three high quality lenses that cover the 63 degree angle of view (35MM equivalent), I thought I’d make an attempt at pulling together a survey.

The three cameras used for this survey:

  • Fujifilm GFX 50R with the GF 45MM F2.8 lens
  • Leica M10 with the 35MM Summilux FLE lens
  • Fujifilm X-Pro2 with the XF23MM F2 lens

Survey context

Before we get into the survey, I want to be clear that this is not a scientific survey; a scientific survey in my view is one that can be consistently replicated by third parties. I think there are too many variables in this survey for that to be possible. My choice of using a real life scene rather than a studio environment would on its own make it very difficult for a third party to replicate.

I chose a real life scene rather than a studio environment because, when I look at the images hanging on my walls, I don’t have any of a studio scene or of test charts. I also believe that there are people and sites online that have already done incredible work (Jim Kasson’s blog as an example) with studio scenes and test charts; I didn’t feel I’d be adding anything of value by doing the same.

I also wanted a scene that was interesting to look at, so I chose Jinshanling in Hebei Province. It’s part of the Great Wall of China and is about a 2.5 hour drive from Beijing city. To reach this particular viewpoint, you need to walk/climb ~8km from the base. It was rather cumbersome to take three camera systems and a tripod up there, but it was ultimately manageable.

Image normalization

I know that people will challenge everything done to normalize the images, so the best I can do is to be transparent of the changes made to the images, so that the viewer can make their own judgement on the validity of the survey and its results.

RAW images are commonly called “digital negatives” which implies that they all start from a base state that is common across platforms and sensor sizes. Unfortunately, this is not the case, which makes conducting a survey such as this very challenging. By their very nature, when you convert a RAW file into an image format that is acceptable to the end user, it has to be manipulated and interpreted by software.

This makes it nearly impossible to normalize a set of images across platforms and sensor sizes. I will therefore provide you with the changes made to the images in order to get them close enough for comparison purposes and equally important, so that you can replicate the test with the same software that has been used here.

The three images in the survey have had the following process applied:

  1. Aperture was normalized according to the common equivalence for depth-of-field and diffraction effects. The GFX image was taken at F11, the M10 at F8, and the X-Pro2 at F5.6
  2. ISO was set at 200 for all cameras which is the base ISO for the M10 and X-Pro2
  3. Focus was set at the front part of the bridge (please see the image below with the yellow “X”)
  4. The three images were imported into Capture 1 Pro 12.0.2
  5. Provia colour profile was applied (The M10 used Scotty Wang’s profiles, Fujifilm cameras used the built-in one)
  6. Auto-adjust was applied to Exposure and Dynamic Range
  7. Levels were auto-adjusted by individual colour channel in order to remove colour casts
  8. Shadows were lifted on all three images to bring out some detail in the dark areas of this very high-dynamic range scene (+9 on the GFX; +10 on the M10; and +22 on the X-Pro2)
  9. Sharpening was left at the setting that C1 applied upon import (GFX was 140,1,1; M10 was 130,0.8,1; and X-Pro2 was 140,0.8,1)
  10. Images were exported as TIF files into Affinity Photo
  11. Images were aligned (best attempt) with the lowest resolution (M10) image becoming the baseline
  12. There was a surprising difference in field of view between the three lenses, so some cropping was required with the final resolution being 5804px x 3953px; the GFX obviously had the most resampling, with the X-Pro2 having a slight amount, and the M10 having none
  13. One image had a person in it which would easily distinguish it from the others; the person was cloned out and does not materially impact the image; the image region impacted covers 150px wide by 212px high, or 0.13% of the total image
  14. Each image was placed in a layer in Affinity and exported as JPEGs with 70% quality, sRGB colour space, and resampled to 5139px × 3500px; this was done to meet the survey software’s image size limits and to respect people’s bandwidth costs
  15. The Affinity Photo file was saved with the layers named the same as the JPEG filenames and the camera used so that it can be shared in the future
Auto focus was used with the two Fujifilm cameras; the LCD with focus peaking was used on the M10 at F1.4 and then stopped down to taking aperture.

Take the survey

March 22, 2019 update: With 505 responses received, the survey is now closed.

If you took the time to read the above, I wanted to say a big thank you to you. A lot of work has gone into producing this survey and I hope that it produces something of value. Please take the survey at this link: https://www.surveymonkey.com/r/LSBZYS2

Once you’ve taken the survey, I’d really appreciate hearing from you in the comments section below. You don’t need to sign-up or log-in to post.

Survey results

Please visit this post for the survey results.

Comments (3)

John, you have experience designing lenses? How much “easier” is it, and why do they cost so much more?

In my experience sensor size matters only in that it is easier to design a performant lens for a larger sensor.

Leave a reply