February 2012 - Digital Group - How Many Pixels?

Meeting Notes March 2009 to 2018.
Post Reply
Posts: 141
Joined: Tue Mar 25, 2008 7:04 pm

February 2012 - Digital Group - How Many Pixels?

Post by spb » Thu Feb 09, 2012 4:52 pm

Tony Kaye ASIS FRPS spoke to us on the subject of 'How Many Pixels Do You Really Need?'.

Tony is Chairman of the RPS Imaging Science Group and spoke with great authority and insight. Tony's presentation slides may be found here and I hesitate to try to precis them! However I will try to pull out some key points for those who couldn't attend.

Firstly there are no hard and fast answers to the question posed - it all depends. The human eye is part of a sophisticated processing system including the brain and the fovea where the highest pixel density of the eye is to be found. The greatest resolution of the eye occurs within a tiny angle of only around two 60ths of a degree. Resolving power is not the same as sharpness. This was illustrated by two images, one a high resolution image, and the other, a low resolution image that had been enhanced via unsharp masking. At low magnifications the lower resolution image looked the sharpest, but when zoomed in the extra detail in the higher resolution image was clearly visible, and preferable.

Viewing distance is another important factor and it may be considered unreasonable to strive for an increase in resolution that can only be seen under a magnifying glass. For practical purposes a distance of 14 inches is taken as a typical viewing distance and the optimum distance for viewing 8"x12" prints.

The practical result of Tony's analysis is that at 14in distance, a 6"x4" print requires about 2Mpx, an 8"x12" print requires about 8Mpx and a 16"x24"print requires about 34Mpx. 35mm film scanning requires high resolution because of the magnification factor but exceeding about 4,000ppi is often pointless as you are merely scanning grain and not the picture information that it conveys. This equates to about 21Mpx. For the purposes of projection, 2Mpx is enough.

We have talked before about the resolution of desktop photo printers such as the Epson range and Tony had some surprising conclusions. My information that the maximum resolution of these printers is 180ppi only seems to be partially true. Tony had high quality gloss prints at 150, 300 and 600 ppi and it was true that we could not tell any difference between the 300 & 600 ppi images by eye alone. However by scanning these prints and then enlarging them, he was able to demonstrate that the level of detail increased with each increase in resolution. He recommended that images are re-sized in Photoshop to exactly 180, 360 or 720 ppi, for Epson Printers (at the chosen size) for best quality, rather than have the printer driver do this by some unknown method.

Tony will do some new tests at 180, 360 & 720 as a result of the excellent discussion last night though he is not expecting anything significantly different. For those with Canon, HP printers etc. a different value from 360 may be appropriate. You can help determine this via a specialised print manager/driver such as Q image. This is what he used to determine that his printer had a max of 720! It’s here http://www.ddisoftware.com/qimage/downloads.htm

Tony went on to talk about raw converters and how they can produce different results in terms of colour accuracy and resolution. His conclusion was that Adobe Camera Raw (as found in Lightroom, Photoshop and Photoshop Elements) is as good as any, and better than most. On the subject of colour spaces, Tony's practical preference is for Adobe 1998, however he recommended sRGB for everyday use and especially for web, screen and club projector. Extreme spaces such as ProPhoto must be used with great care and only in 16bit mode. There will be dire consequences on image quality if these are not converted to sRGB in images for club use!

All of this excellent analysis and best practice can be laid waste by other factors such as camera shake, lens aberrations and diffraction. On the latter point, the smallest apertures should only be used if depth of field is more important to the image than ultimate sharpness. Sensor noise is another source of loss which increases with pixel count and with decrease in sensor size. There was some amusement at the prospect of the Nikon D800 announced this week with no less than 36Mpx in an FX (ie 35mm sized) sensor. Although this camera could have applications in studio and other specialised tasks, Tony advised that nothing but the finest lenses should be used with it to see the benefit. This was even more true of the D800E version with no anti-aliasing filter which can resolve greater detail at the expense of aliasing and moire patterning. We would be best advised to spend our money on better lenses rather than bodies with more pixels. Cameras with 6 million 'good quality' pixels can make better prints than cameras with 16 million small pixels, poor optics and aggressive noise processing and compression.

Finally we came to an answer to the oft-asked question "How many pixels is equivalent to colour film?" After many qualifications, Tony plumped for a figure between 20 and 25Mpx, though the number you get really depends upon what assumptions you make.


Thursday 1st March 2012

Cheers, Steve.

Posts: 141
Joined: Tue Mar 25, 2008 7:04 pm

Re: February 2012 - Digital Group - How Many Pixels?

Post by spb » Fri Feb 10, 2012 3:33 pm

Here's a subsequent update from Tony;

Thanks for the link. I have done some work and it really confirmed my expectations. You may want to share it:-

1) I took a RAW file through ACR at 720 PPI and its “size” was approximately 4in x 6 in.
2) I downsampled the 720 PPI file using bicubic sharper to 360 PPI still a 4 x 6.
3) I downsampled the 720 PPI file using bicubic sharper to 180 PPI still a 4 x 6.
4) I printed all the files on my Epson R265 printer (pretty low end but still a photo printer!) with the driver set to Epson Premium Glossy “Photo RPM”
5) I printed all the files on my Epson R265 printer again with the driver set to Epson Premium Glossy “Photo”
6) All prints were scanned in at the maximum optical resolution of my aged Epson scanner, 1200 PPI

The attached file is a composite and it shows a very similar result to the 600/300/150 prints I showed last night:-

a) At “photo RPM” the most detail is in the print from the 720 PPI file, followed by the 360, followed by the 180. This suggests to me the printer is really running at 720 PPI.
b) At “photo” the difference in detail between the 720 PPI file and the 360 PPI file to my eyes is no longer detectable, and the 180 still has the least detail. Additionally the level of detail seen from the 360 and 720 PPI files at “photo” is similar but noisier than the print from the 360PPI file printed at “photo RPM”. My conclusion is “Photo” is 360 PPI.

When viewed by my eyes I cannot see any difference in detail between the prints from the 720 and 360 PPI files irrespective of which driver setting was used, but the 180 looked a little less detailed when viewed under good lighting. Also the scans show that one of the nozzles on the printer may be blocked!

Attached are sections from the scans. Tell us what you think!

Best regards


NB I've had to more heavily compress the following image to post it on the forum but I think you can still see the effect. Let me know if you want to seen the original. Steve.
Printer-Resolution2.jpg (96.41 KiB) Viewed 5138 times

User avatar
David P
Posts: 4
Joined: Tue Mar 25, 2008 8:58 pm

Re: February 2012 - Digital Group - How Many Pixels?

Post by David P » Fri Feb 10, 2012 7:00 pm

The Epson spec for Tony's R265 is 5760x1440 dpi.
Presumably, maximum effective definition will occur when the printer can fit an exact number of dots into each pixel. Hence the recommendation to print at 720ppi (2 dots per pixel in one direction), 480ppi (3 dots), 360ppi (4) and so on. If you print at, say, 400ppi then the printer has to fit 3.6 dots into one pixel and there will inevitably be some loss of definition along the edge of the pixel. Interestingly, the usually demanded 300ppi, will not be optimum according to this theory.
I guess that RPM on Tony's printer uses the full 5760x1440 resolution, while Photo only goes to, say, 2880x720. 1 dot per pixel and 2 per pixel are both so coarse that you may not see any difference.
But, as Tony said, 'What the eye doesn't see . . . '


Post Reply