This brings up one of my favorite quotations:
Lord Kelvin [PLA, vol. 1, "Electrical Units of Measurement", 1883-05-03]
"To measure is to know."
"If you can not measure it, you can not improve it."
"In physical science the first essential step in the direction of learning any subject is to find principles of numerical reckoning and practicable methods for measuring some quality connected with it. I often say that when you can measure what you are speaking about, and express it in numbers, you know something about it; but when you cannot measure it, when you cannot express it in numbers, your knowledge is of a meagre and unsatisfactory kind; it may be the beginning of knowledge, but you have scarcely in your thoughts advanced to the state of Science, whatever the matter may be."
I think that quantitative and qualitative (subjective) testing both have a place. Indeed two of my favorite reviewers (Bjorn and Diglloyd) both do qualitative testing. However, quantitative testing helps us to understand what parameters are really important to image quality and how IQ is subject to the laws of physics. For example, the statement by Michael that MFDBs have six stops more DR than dSLRs in untenable from a scientific viewpoint as explained in detail by Emil Martinec, who is a highly published physicist at the University of Chicago. Emil concludes:
"According to the engineering standard the MFDB is still well short of the D3x, however according to a standard more relevant to photography the back comes out slightly better (but less than 1/3 stop), mostly because of the larger sensor area collecting more light over the frame. The difference is not however the many stops DR advantage that some MFDB proponents claim."
I can already anticipate Jeff Schewe's comment that Emil has likely never used the Phase 1 D65+, so his opinion should be discounted.
What the Phase One does have is megapixels, but it is not clear that those pixels are superior to those of a good dSLR such as the Nikon D3x. Diglloyd (a pay site, but well worth the modest fee), compared the D3x to the Leica S2 and concluded that their per pixel performance was similar if deconvolution sharpening was used with the Nikon to counteract the softening produced by the blur filter.
The P65+ may have 16 bit output, but with an engineering DR of 11.51 stops and a tonal range of 9 bits (DXO), those extra two bits are wasted recording noise. The lack of a blur filter improves apparent sharpness but can result in alaising. Why does Nikon go to the expense of using a blur filter (as does Canon and almost all other dSRL makers)? I understand that a blur filter would be prohibitively expensive for a MFDB, so it is really not an option there.
Regards,
Bill
Bill,
Good post! We sometimes need to be reminded of these basic scientific principles of observation and measurement which many people still seem to ignore even centuries after they were first espoused, and despite their having benefited so much from the fruits of such applied principles.
I was a bit surprised at Jeff Schewe's comment implying that one needs to
use an MFDB before one can get an idea of the quality of its output. In fact, the remark is quite insulting, implying that some of us may not be smart enough to deduce and imagine the improved image quality one might expect from a larger sensor with a higher pixel count, especially considering the wealth of measurement that is published at DXOMark, which can inform and modify a subjective impression or deduction.
An example of deduction, for Jeff's benefit: When my best DSLR was the 8mp Canon 20D, I bought a Canon 24mm TS-E, mainly for the purpose of stitching, in those days when stitching programs were not nearly as good as they are today.
I have a number of stitched images from that lens and camera, some of which I've printed big (like, 24" x 40" - not big by Jeff's standards, but appropriately big for the file size) and which hang on a wall. I'm impressed with the clarity and the detail and the subtle shading that is apparent from a close inspection.
On such occasions I remind myself that this is the sort of quality I could expect from a single shot with a 5D2, cropped to the same aspect ratio (except when the camera is horizontal and the aspect ratio is very wide in the 20D stitch - in which case the 5D2 could not compete as a single, cropped shot).
'But what about the pixel quality of the old 20D?', I ask myself. Perhaps the cropped 5D2 image would have better dynamic range, lower noise, superior color rendition.
Okay! Let's have a look at some
real measurments. I refer to DXOMark's graphs comparing the 20D with the 5D2, and find, surprise! surprise!, that the 5D2 pixel is no better than the 20D pixel. The differences are so marginal, they are of no consequence in practice. (To see this, you need to go into 'screen' mode which compares pixel with pixel).
So there you have it. A simple process of intelligent deduction.
Of course, the 5D2 has a wealth of other useful qualities which the 20D lacks. This comparison relates only to basic image quality under ideal conditions that might allow for stitching.