Why I cannot see a difference in A2 size prints?

Monday, 05 May 2014 00:00 administrator

Note: This article is still in progress, more stuff will be added

I started shooting APS-C at 12 MP got into full frame 135 with 24 MP and now I am shooting also shooting 39 MP MFD. Normally I print in A2 size (16.5"x23"). At that size I can observe little difference or none at all between my cameras. There are huge differences in the files, but little in print. Why is this so?

Looking at SQF 

Ed Grainger at Esatman Kodak developed something called Subjective Quality Factor. This is based on human vision and measures the perceived image sharpness. More exactly it is the integral of the contrast sensivity function of the human eye at a given viewing distance. SQF is a well established base for comparison.  A good description of SQF is given here another less scientific one is here.

I downloaded a couple of lab test charts from DPReview and came up with this table.

CameraSQF (LS)SQF (SS)Comments
Sony A799.198.2
Sony A7r99.598.8
P45+ (my image)93.698.2
IQ180 + Schneider 80/2.896.298.8

The left column is with Standard "landscape/narrow edge" sharpening while the righ column is the values imatest calculates with "standard sharpening", I think that column is most relevant.

Normally it is said to take about 5 points of difference in SQF for a visible difference. Looking at the column at standard sharpening the values are very close.

Why did I use DPReview images instead my own? Two reasons DPReview tests more cameras than what I have, but also that it has been suggested that my technique using MFD is not proper. Using test images from a well know test site should eliminate bad results caused by my technique. 

Looking at real prints

The "DPReview" test target is well thought out, I cropped a central portion down to 50% and printed in A4 size. This corresponds pretty exactly to A2 size. I tried to look carefully at the images. Could I see a difference? Yes and no!

The areas where I could se difference in the print was where there was obvious aliasing, and that was the resolution trumpet and the fine text parts. Especially on the left resolution trumpet aliasing was very obvious on the A7 but not at all on the A7r and the IQ180. Readibility on fine text was very much better on the IQ180, but the difference was only observed when viewed close (like 25 cm). 

Normal structures were very similar in print, so I would not be able to tell the images apart.

Human vision
Human vision can normally disolve around one minute of arc, 1/60-th of a degree. This resolution is called 20/20 vision. Vision is normally measured by the so called Snellen chart. The Snellen figures cover 5 minutes of arc. 

In general it is normally regarded that 180 PPI is needed for a very good print. If we choose 50 cm viewing distance and calculate 500 * sin(1/60) we get 0.145 mm or 175PPI. So , the normal requirement for a very good print corresponds to the resolution of human vision at 50 cm. It is regarded that the maximum resolution of human vision is achieved at 25 cm, and that would correspond 360 PPI, exactly the normal resolution of Epson printers.

How come that the 180 PPI figure is regarded adequate? One possible explanation is that although human vision can resolve 1/60 of a degree, the maximum contrast sensitivity of the eye is around 5 cycles per degree, or 6 minutes of arc. So vision can detect small detail but it is most sensitive to much coarser detail.

On the other hand, there is something called vernier acuity. human vision can see some defects well beyond the normal resolution of vision. A good example is discontinous lines, which are clearly visible even beyond normal resolution of vision.

Discussing some detail in the DPReview images

I made prints from the images, but they are hard to post over the net. So I made the follwing excercise:

DPReview test images from IQ-180 (80 MP MFD), Sony A7r (36 MP FF 135 camera) and Sony A7 (24 MP FF 135 camera) were used. The two former cameras lack OLP filter while the A7 is OLP-filtered.

All images were dowscaled to 40 cm image height at 180PPI. Four parts were taken out

The left hand resolution trumpet

In this part of the image it can clearly be seen that the IQ-180 offers some advantage, but the A7 image shows clear artifacts. In this case the subject/lens outresolves the sensor creting aliasing artifacts. The OLP filtering is not strong enough to significantly reduce monochrome aliasing.

The Beatles group

In this image the resolution advantage of the IQ-180 is far less visible.

The hairy balls

Although this part of the image contain fairly fine high contrast detail the resolution advantage of the hig res cameras is not very obvious.

High contrast text

In this case the resolution difference is visible. What is most visible is the aliasing of the A7, which breaks up the contours of the text and shows some fake colour.

Note: most computer screens have around 100PPI while these images are at 180 PPI. So looking at normal computer screen is like looking at the A2 size images at 28 cm.

My conclusion is the there is some advantage to higher resolution but it would be more visible on test charts than real world subjects.

A real world sample

The two samples below were shot with a 24 MP DSLR (Sony Alpha 99) and a 39 MP MFD (P45+). The 24 MP image was upsized to match short dimension on the 39 MP image. The 24 MP DSLR image (on top) is clearly inferior to the 39 MP MFD image (below). These two images were cropped down to half size and the cropped images were printed on A4 paper at 720 PPI on an Epson SP3800 printer on semigloss paper.

As the images were cropped to half size and printed at A4 size the full size image would correspond to A2 size, around 16.5"x23". The two prints were essentially identical when examined by the naked eye. I also consulted a friend with considerable experience in professional printing, he was once working at a professional lab. He actually found the DSLR image slightly superior.

Analysing the prints with a 5X loupe it was very obvious that the the MFD image had superior detail, but without the loupe the images were impossible to tell apart. Well, almost impossible. Making two identical prints from two different images is very hard.

As a side note, these images were published on Luminous Landscape in this thread. On the LuLa thread I got some significant bad critique for bad technique, which may be relevant or not. That critique missed the important point, however, name that the DSLR image was markedly inferior at actual pixels at sime size, but those differences were not visible in a A2 size print without a loupe. That means that the 24 MP digital is good enough for A2, and going to higher resolution gives very little benefits as they are not clearly visible in A2 size prints. 

Obviously, the situation would differ at larger size, I made larger size prints initially by mistake, and there a difference may have been visible.

Very obviously, both images could probably be improved my more advanced sharpening, but any such sharpening could be applied to both images.

Using test targets vs real world subjects

I feel it is safe to state that test targets give a better indication of resolution differences then real world subjects. The main reason is there is only one plane of optimum focus. A real world subject is seldom flat and parallell with the sensor, so much of a real subject will probably be slightly out of focus. So a test target essentially is the optimum case where differences most likely will be visible.

That also means of course that out of focus rendering plays a major role, but out of focus rendition is more of a property of the lens than of sensor size or sensor resolution.

Summary of findings

The only conclusion I can draw from the images I have tested is that a 24 MP DSLR with a decent lens can produce images good enough for A2 and probably larger. Going to higher resolution will probably not yield a visual improvement in A2 size prints at any viewing distance with normal vision. Here it needs to be taken into account that the images were studied by two persons at 50+ age. Younger eyes may be more sensitive to the differences. Both test persons were near sighted.

These finding are pretty consistent with SQF data.

Last Updated on Thursday, 08 May 2014 20:41