Digtal vs 8x10 film

[andylynn]

<p>That Phase One's sensor is smaller than a 645 frame. If you do the math it's at about 4900 PPI, so smaller pixels than my D7000. You do get to use Mamiya glass, which is nice, and 80MP is a lot, but an 8x10 frame has 23x as much area as this camera's sensor, and you can easily get Ektar, Portra, Velvia, Delta, etc. in 8x10.</p>

<p>All they've done by running the scans at low res to match the pixel count of the digital, not using proper sharpening, and doing a sharpness comparison, is to show that bad scanning technique kills sharpness.</p>

[larrydressler]

<p>An 8x10 contact print from Tri-X shows no grain... Or pixels or noise or whatever you want to call it even if you have to do digital on a V700... LORd... I want to quit.... even at 300 it will overload most computers if you want to tiff it.</p>

[chrisnielsen]

<p>That whole test is unbelievably retarded. At least silly old K.R. comes out and says up front not to take him seriously</p>

[jason]

<p>I found the article in LL fascinating. Imagine comparing a poor scan to a Hi-Res. Digital. Methinks my Canon Super-shot S590 might indeed be better than 8 x 10. My printing needs are smaller. Mostly 4 x 6, occasional 8 x 10/12. If one was to "optically"print the large format..<br>

Ken Rockwell "tests" are great! Everything there is a fair comparison. Folks who bought into the Leica M8/M9 must really be mad when an ancient 1956 Leica M-3,60's Summicron 50mm, sails past the digital M9/M8 in terms of color and feel using Velvia. Film has this wonderful "organic" look. Digital is way more precise but cold..<br>

I use film and digital. Digital allows me to shoot till I drop. Film makes me question each exposure. This leads to poorer photos! I am only safe when I use Black and White film because there are no snotty nosed,fat fingered and least caring people in the chain of development and proofing. I am my own lab.<br>

The LL article proved absolutely NOTHING. What did you expect from a site that knew and concealed the major flaw of the Leica M8? The magenta color cast which I get on my really cheap old Kodak Easy Share.</p>

[Dave Luttmann]

<p>Maybe I should go dig up my old D30 vs 35mm Provia comparison. It'll show how they don't know how to scan or PP film at the LL. My favorite though was his 6x7 vs the 1Ds where he said the 1Ds had more resolution. Then a couple of years later, he posted the 645 vs 1Ds2 and said the 1Ds2 had almost as much rez as the 645. Huh? Let's see....11mp better than 6x7, but 17mp worse than 645. Oh well, I just keep laughing at their "tests."</p>

<p>Ya Larry, I know, the contrast would be off the chart pushed that much. My point is that I don't get that much grain from 4x5 FP4 scanned at 3200ppi.....so how does he get that much grain from Acros at 745 ppi....or as he says, "DPI." I've yet to figure out how a scanner "dots" an image....or scans an image into "dots." I guess we have someone who doesn't know the difference between printing and scanning.</p>

[StuartMoxham]

<p>I don't get that much grain from Neopan400 in Rodinal scanned a 800ppi.<br>

The writer posted a follow up article. Someone on rangefinder forum posted a link to this.<br>

http://www.markuszuber.com/8by10.html</p>

 

[mauro_franic]

<p>I like the comment he adds with TMAX 400 would should show even more grain but no more detail at 745dpi.<br>

<br /> The reason why I take the time to post these things is because unscrupulous test like theirs are posted for personal financial reasons at the expense of hurting an entire community of artists.</p>

<p>So here is what film looks like:</p><div></div>

[Dave Luttmann]

<p>Thanks Mauro. This is something many of us have pointed out over the years. I recall posting my D30 vs Provia samples at DPReview about 9 or 10 years ago....where I was labelled a liar, biased, etc, because I simply showed the Provia to far outresolve the D30...and even my 10D. I had many people email me saying they found the same thing...but didn't want to get roasted by all the LL followers.</p>

<p>I can tell you this much...I get better results from my Epson V700 for 4x5 than the LL did in this test with 8x10. That should about sum it up for everyone.</p>

<p>Maybe I should set up something in my blog so it can be referrenced by people who beleive the LL nonsense.</p>

[larrydressler]

<p>:-)</p>

[mauro_franic]

<p>The two crops on top also represent the difference in resolving power between the 80MP MF digital and an 8x10 piece of fast film. </p>

<p>The difference is so meaganormous that the LL's undertaking this project tops their history of embarrasements. </p>

[mauro_franic]

<p>FYI the shot above was shot with my Mamiya RZ67 and scaned at 97 megapixels (4,000 native dpi) with my Coolscan 9000.</p>

[Dave Luttmann]

<p>But Mauro.....how can that be? LL has already said there is no additional information other than grain beyond 745 ppi.... ;-)</p>

<p>This one is easy. If LL is correct, then your 745ppi and 4000ppi images should look the same. If they don't look the same, then LL has some "Splainin" to do!</p>

[mauro_franic]

<p>Well, since they drank their own coolaid and have been shotting landscapes with 3mp to 20mp cameras for the last 10 years, the have no choice but to make up a story so people won't turn their 30x40 prints into torches and come to their doors.</p>

<p>The funniest thing is that Michael R. himself emailed me saying he didn't want to do a side by side testing with me of his 80mp MF and my MF film. </p>

<p>Actually even funnier is that people without experience will quote this test from the Luminous Landscape for years to come.</p>

[Dave Luttmann]

<p>This is even being questioned at the LL. Lenny Eiger is an expert in scanning and says:</p>

<p>"It's just plain ridiculous and the article should be pulled. It's clearly misleading."</p>

<p>I really wish they would stop posting the useless, misleading tests.<br /><br /></p>

[alan_rockwood]

<p>Larry,</p>

<p>If you are not interest in this topic then why don't just you stay away and leave the thread to those who are truly interested in a meaningful discussion on this topic. Just because you don't like the topic does not mean that everyone else needs to share your opinion. Your pejorative comments in this thread are not helpful in any way whatsoever.</p>

[larrydressler]

<p>Alan I am following this with interest.</p>

[leighb]

<p>There is no scanner in existence (that I know of) that can extract the available information from a high-quality 8x10 (or 4x5) image. This can be proven incontrovertibly by the numbers.</p>

<p>Given that a high-quality film and lens combination can resolve 100 lp/mm (the actual number is irrelevant):<br /> A line pair is defined as a black line adjacent to a white line, so you have 200 lines per millimeter.<br /> Looking at this on a single scan line you have 200 dots per mm, alternating black and white.</p>

<p>Standard sampling theory requires that you take at least two samples of each data element*, so a basic scan of that line would require 400 samples per mm.</p>

<p>That equates to a scanner resolution of 10,160 dpi!</p>

<p>Even that resolution won't provide a high-quality data set, only a marginal one. An accurate scan would require four to eight times that resolution.</p>

<p>Numbers don't lie.</p>

<p>- Leigh</p>

<p>*Note: The 2x sampling criterion is universal in the electronic world. It applies to all types of data acquisition. It has nothing to do with photography.</p>

[mauro_franic]

<p>Leigh, no one questions that the film can capture at least 20 times the detail than the 80MP digital camera.</p>

<p>Even with my Mamiya 6x7 cameras and Coolscan I can get almost 100 MP of detail. The limitations on my workflow (in order) are:</p>

<p>1- the scanner at aprox 150 lpmm,<br /> 2- followed by the film (Velvia or Ektar: 160 lpmm, Tmax-Xtol: 220 lpmm or TPan-Technidol: 300 lpmm),<br /> 3- and lastly the lenses - I have tested my mamiya lenses aprox 400 lpmm (measured on microscope projection).</p>

<p>The Luminous landscape is only showing the lowest common denominator of their lens, technique, focusing and laughable scanning resolution. It is worth noting that their results are so extremely poor that cannot be attributed to incompetence or equipment; they clearly show intention to mislead.</p>

[mauro_franic]

<p>By my previous crops you can appreciate the vast difference in resolving power between film and the 80mp MF camera. That's why the test is a joke.</p>

<div></div>

[mauro_franic]

<p>Now scan only at 745dpi and don't sharpen and you get the result they presented (assuming their technique for the shot was good).</p>

<p>Luminous Landscapes 745 dpi with improper/no sharpening:</p>

<div></div>

[alan_rockwood]

<p>Leigh,</p>

<p>I agree with your analysis conceptually, but i believe there is in error by a factor of two. The Nyquist sampling theory says that the sampling rate needs to be at least 2X the frequency of the highest frequency component in the signal in order to reconstruct the signal without error. If we pretend that a white/black pair corresponds to one period of a sinusoidal signal (a slight simplification, but not too far from wrong) then the Nyquist limit is two sampling points per black/white line pair. This is a factor of two lower sampling than the estimate you provided.</p>

<p>As a practical matter if you are close to the Nyquist limit you could end up with some kind of Moire pattern if the signal is an extended sinusoidal signal, so it is better to sample at a rate somewhat higher than the Nyquist limit.</p>

 

[leighb]

<blockquote>

<p>If we pretend that a white/black pair corresponds to one period of a sinusoidal signal...</p>

</blockquote>

<p>Hi Alan,</p>

<p>That's the error. Each line is a full cycle, not a half cycle.</p>

<p>This is easily demonstrated by shifting the sample point relative to the data set.</p>

<p>If you have a sensor with 200 elements/mm, corresponding to the 200 lines of the original data, and you align the sensor so each element lines up exactly with the center of each line, you can acquire the data as alternating black and white.</p>

<p>However, if you shift the sensor by one half line width, you end up with a data set that is uniformly gray, each sensor element seeing one half white and one half black in the subject.</p>

<p>Since the alignment of the sensor with the original is completely random, you must treat each line as a separate full cycle.</p>

<p>Thus my original 400 sample value is correct.</p>

<p>- Leigh</p>

<p> </p>

[leighb]

<blockquote>

<p>It is worth noting that their results are so extremely poor that cannot be attributed to incompetence or equipment; they clearly show intention to mislead.</p>

</blockquote>

<p>Hi Mauro,</p>

<p>I agree completely. I have not read the original article, and don't intend to do so.</p>

<p>It seems obvious from the comments here that LL has an agenda and is trying to fabricate some 'science' to back it up...</p>

<p>unsuccessfully, but that won't prevent its propagation in internet history. ;-(</p>

<p>- Leigh</p>

[steve m smith]

<p>The way to be sure would be to have the best scan possible made from an 8x10 negative then downsize it to 80 megapixels. If there is any difference then there is clearly more detail in the film.... Or is that what Mauro did?</p>

[alan_rockwood]

<p>Leigh,</p>

<p>Consider a simplified example. Assume there is one line pair per inch. That would be 1/2 inch of black and 1/2 inch of white. The spatial frequency of this signal is 1.l (To be strictly correct and include the units the spatial frequency would be 1/inch.) The Nyquist limit is to sample at twice the spatial frequency, i.e. to sample the signal at a rate of two times per inch. Thus, there is one line pair per inch, and you must sample at two samples per inch. Everything scales linearly with frequency, whatever the spatial frequency is (in line pairs per inch), so the Nyquist limit is twice the number of line pairs per inch. According to the Nyquist sampling theory, if the signal is bandwidth limited such that the highest frequency in the signal is less than the Nyquist limit then you can mathematically reconstruct the original function without error if you sample at the Nyquist sampling frequency.<br>

<br /> Your comment about the strange results that you would get by a slight shift of the sampling grid (your comment about uniformly gray) is correct, i.e. you would get a uniformly gray result. That is a special case of the Moire pattern problem I mentioned, and it is a good reason for sampling at a rate that is much higher than the nominal Nyquist limit. It is a subtle and interesting point, but nevertheless the Nyquist limit is twice the spatial frequency.</p>

<p>Looking at the Moire problem a little deeper, suppose you were sampling at a rate that is just a little higher than the Nyquist limit, such as 2.02 samples per inch if we had a signal with a spatial frequency of 1/inch (i.e. one line pair per inch.) You could start off being in phase with the signal, i.e. the sampling points being aligned with the light and dark parts of pattern. In that case you will capture the alternating light/dark pattern in that part of the image. However, after proceeding part way across the frame the sampling device (the image sensor) would get out of phase and it would sample the nodes in the pattern, i.e. the region right at the light/dark boundary. During this part of the image you would get a uniformly gray result rather than a light/dark pattern. (This is similar to the ) Proceeding still further across the frame, the sampling device would again be in phase with the image and reproduce the alternating light/dark pattern. Thus, sampling at close to the Nyquist is not actually good enough to do a good job of capturing an image, and in that sense I agree with your comment about sampling at a higher rate. In the specific example you cited you recommended sampling at twice the Nyquist limit, and I agree that it is a good idea.</p>

<p>I think one of the implications of this discussion is that you need to sample at a rate substantially higher than the Nyquist frequency if you are going to produce a direct image. However, if you are going to mathematically process the sampled signal and then reconstruct the image then sampling at the Nyqist sampling rate (actually, just a slightly higher frequency than the nominal nyqist limit) would be enough to perfectly reconstruct the image.</p>

<p>At least that's my analysis. It doesn't necessarily mean I am right, but I have been using Fourier transforms in my work for many nears, least some degree, and I have published several papers based on Fourier analysis of certain functions, and I am therefore pretty confident in my analysis.</p>