My guess is that we see is that the human vision is most sensitive to medium frequency detail.
Hi Erik,
That's correct, but it is very much connected to things like output sharpening and printing at 600 or 720 PPI.
The eye can resolve fine detail but medium frequencies will dominate perception, see linked figure from Wikipedia. So if we discuss detail visible at 360PPI at 25 cm, the contrast sensivity of vision is several magnitudes below maximum.
Yes,
unless we boost the contrast of the lowest spatial frequencies of the output image!!! Then we do not only see a bit more micro detail, but also the spatial frequencies from medium frequencies to those highest spatial frequencies will receive more signal/contrast. That would partly compensate for the dropping sensitivity of human vision, and the
combined (boosted)
input signal x CSF = output signal drop will be slower. The image subjects will look more 'tactile' as a result. Having the additional pixels at 600 or 720 PPI, helps in boosting the signal even more, without creating artifacts that would be more likely to become visible with only 1/4th of the pixels to calculate with.
That is separate from all the other cognitive processes involved in 'vision', but in a double blind test when specifically scoring on perceived detail, it should make a difference.
To me that indicates that we should focus our sharpening efforts more on medium frequencies than actual pixel detail.
The problem with that is, that we usually cannot control the viewing distance from the output. So what may be medium spatial frequencies at a proper viewing distance that allows us to take in the entire image/composition without turning our heads too much, will become low frequency detail, and high frequency will become medium frequency detail, when we get closer. That's why we need to
also sharpen the higher spatial frequencies, and that will automatically also start boosting the medium frequencies. So that makes the image quality less sensitive to different viewing distances. It's like boosting the MTF curve near the Nyquist frequency, it will lift all spatial frequencies below it as well, but in a controlled gradual/monotonic way.
Figure 24 at
this site shows that the contrast sensitivity for the spatial frequencies also depends on the average luminance level or viewing conditions. Spatial frequencies of 60 cycles/degree (= average 20/20 vision, younger people can do better than that) would be around 300 PPI at reading distances, and are worthwhile to boost if we want to exploit that. We can do it better and with fewer artifacts when we boost the levels with even finer detail, say 600 or 720 PPI, and the lower frequencies will smoothly follow as well.
BTW, as cited in the link above, the receptor array in the human visual system can resolve in the order of 6/1 (20/3) or ~150 cycles/degree, which one might add would roughly match the 600-720 PPI at reading distance range. However, the rest of the eye's structure (a.o. lens) will reduce that to closer to the earlier mentioned 60 cycles/degree.
The impression I got was that the visitors placed that image #2 of all pictures at the exhibition.
That may well be because they loved the scene (as they should), more than they scored it on perceived resolution, crispness, 'tactile' quality. But those qualities do help to enhance the immersive quality of being there, if done well.
At the risk of repeating myself too much, Topaz Detail does just that, extremely well (although it gets slower with all the calculations needed as sizes increase), and it not only allows to control the small details but also those 'medium' detail frequencies separately for even more control. Detail can even be locally brushed in (or out) with edge aware masking, to avoid sharpening the noise in e.g. smooth sky gradients. Every enhancement smoothly transitions between different spatial frequencies, and can be finely controlled for overall tonality, and/or additionally for shadows and/or highlights.
Cheers,
Bart
P.S. In the book,
A System Engineering Approach to Imaging, By Norman S. Kopeika, there is a chapter on the "Threshold Contrast Curve". I've attached a chart from that chapter. It basically shows, translated to our print resolution challenge, that we need to boost the contrast of the highest spatial frequencies of the input signal (towards an MTF=100%) more, to offset the loss of contrast sensitivity of the human visual system's MTF . It basically peters out at 60 cycles/degree, but Nyquist taught us that we need more than twice that sampling frequency to reliably resolve (and edit) those frequencies.