Please lift me out of my ignorance - scanner resolution

[david_goldenberg1]

<p>The "resolution" that manufacturers specify reflects only the number of pixels that are generated in the output files, i.e. the Nyquist-limited resolution (though even that can be fudged by using interpolation). These specifications do not say anything about the ability of the scanner to actually resolve features on the negative or reflective medium. <br>

For those who are interested in a serious evaluation of scanner resolution, I suggest the following paper:<br>

http://www.aspbooks.org/a/volumes/article_details/?paper_id=30173<br>

The author Robert Simcoe is an engineer who recently led a group building a specialized high-resolution scanner to digitize astronomical photographic plates. I think it is reasonable to assume that he knows what he is talking about. In this paper, he compares an Epson V750 and Nikon Coolscan 9000, using a slanted-edge method to measure the modulation transfer function (MTF). In brief, the resolution of the Coolscan approaches its advertised Nyquist resolution (4000 ppi), but the Epson falls far short of its advertised resolution. This is almost certainly due to the optical limitations. In addition, Simcoe explains how the advertised resolutions of flat-bed scanners are "padded" by interpolating data from staggered CCD arrays. This also contributes to the difference between the specifications and real resolution.<br>

I have also used the slanted edge test to measure the MTF of a Coolscan 8000 and an Epson V500. I didn't do it as carefully as Simcoe did, or with as good a target, but my results were quite similar to what he describes.<br>

The reason that Epson and others can get away with the specifications they cite is that the numbers have a carefully-specified meaning, it's just not what most people would expect "resolution" to mean. Personally, I think that this is unfortunate. The real resolution of inexpensive flat-bed scanners is really very impressive for their cost, without any spec-manship. Not surprisingly, though, the resolution is not comparable to much more expensive film scanners.<br>

I have a copy of the paper described above, but it is copyright protected. If you send me a personal message, I will send a pdf to you (assuming that I don't get hundreds of requests!).<br>

David</p>

<p> </p>

[zack_zoll]

<p>David, this spec discussion reminds me very much of how cheap speakers are rated. Generally speakers are rated in wattage (how much power they can handle), and for the better ones, sensitivity (how loud they are per watt). An inexpensive car subwoofer would very frequently be advertised at 1000 watts. But if you played a song into it from a 1000 watt amplifier, the speaker will blow in seconds. They get the 1000 watt rating because it can handle one very specific frequency at 1000 watts, and not because you can actually pump 1000 watts into it. The actual watt handling might be around 150.</p>

<p>Isn't marketing great?</p>

[mattb1]

IMO Epsons are really only able to capture 2400 dpi, expecting more is just setting yourself up for disappointment.

Why, just as some have said: lens, light surce, ccd, stepper motor, film holders, and scanning through glass. Why can

they claim higher? Because they can give you a file of a higher "resolution", so they meet the truth in advertising idea.

No where does it say that those higher "resolution" files have a one to one relationship between a pixel and a sample

site.

 

Should you buy Epsons? That is up to you, if the tool fits your needs then it might be ok to use it. Be careful in

evaluating examples. I found that after being on the scanning learning curve for a little while that I became aware of the

effects of over post processing of images, I missed that a lot of examples were not living up to the standards I wanted

to achieve. So, I bought wrong and wasted a good amount of time.

 

Of course take my opinion in it's contex, I wanted to get all the detail I could out of the film I shot. To be frank the Nikons fall a little short in that regard as well, but still more than twice as good as the Epsons.

[marc_bergman1]

<p>David,</p>

<p>Thanks for posting the link. I downloaded the article. It gives the best explanation for the lower resolution found on flatbed scanners that I have read.</p>

<p>This is the type of information that enhances our understanding of the scanning hardware available. I think this can aid buying decisions that match the user's needs and budget.</p>

<p> </p>

[richardsperry]

<blockquote>

<p>IMO Epsons are really only able to capture 2400 dpi, expecting more is just setting yourself up for disappointment.</p>

</blockquote>

<p>Scanning at 6400 and 2400 respectively produce two different apparent images, the 2400 has larger pixels at the same size magnification.</p>

<p>Also, if what you say is true, then all that is happening is that the scanner is splitting the same pixel to make more than one, say 4 for convenience sake. Magnifying down to say a block of 4 pixels in the 6400 scanned file, should reveal that the pixels are the same color. And that does not occur.</p>

<p>Additionally, if you are repeating the German's conclusion that one should limit scanning dpi to 2400 with the Epsons; one should keep in mind post processing(who would print a raw scan?). For example, Unsharp Mask a 2400 scan and a 6400 scan at the same setting, or even converted/comparable setting, reveals two very different results.</p>

 

[richardsperry]

<p>Here is a 35mm scan, on the Epson V500(at $160), at the PNet max size of 700px inline.<br>

<img src="http://img135.imageshack.us/img135/4526/skiliftedit.jpg" alt="" /></p>

<p>Here is a 100% selection for comparison(700x700).<br>

<img src="http://img839.imageshack.us/img839/3858/expertsonly.jpg" alt="" width="700" height="700" /></p>

 

[marc_bergman1]

<blockquote>Additionally, if you are repeating the German's conclusion that one should limit scanning dpi to 2400 with the Epsons; one should keep in mind post processing(who would print a raw scan?).</blockquote>

<p>Richard,</p>

<p>I couldn't find where the German site specifies that one should limit scanning to 2400dpi on any of their Epson scanner tests.</p>

<p>Typically in their tests they make resolution tests at various settings to see if there is any improvement in using a higher setting. If all you get is a larger file then they recommend using the lower setting. This is what happened with their Epson V600 test. They found that using the 3200ppi setting gave the maximum resolution along with a smaller file size.</p>

[richardsperry]

<p>Marc,</p>

<p>I get a much better scan at 6400dpi at a lower percent size, say 25%, than with lowering the dpi; while decreasing scan time and file size. With the V500; considering that the V500 and V600 have similar specs, I will assume that they are scanning relatively the same.</p>

<blockquote>

<p>I couldn't find where the German site specifies that one should limit scanning to 2400dpi on any of their Epson scanner tests.</p>

</blockquote>

<p>I was in error, this is what he wrote.</p>

<blockquote>

<p>In order to achieve the maximum resolution of 1560ppi, one does not have only to scan the with the highest optical resolution but it is sufficient to digitalize the original with 3200ppi. The effectively achieved resolution does not differ, no matter if one scans with 3200ppi or with 6400ppi - in both cases, an effective resolution of 1560ppi will be achieved! By this way, it does not make sense to scan with the Epson Perfection V600 Photowith 6400ppi.</p>

</blockquote>

<p>From, http://www.filmscanner.info/en/EpsonPerfectionV600Photo.html</p>

[leighb]

<blockquote>

<p>...in both cases, an effective resolution of 1560ppi will be achieved!</p>

</blockquote>

<p>Given that they only wanted 1560 dpi in the first place...<br>

how could their opinion possibly be relevant to the maximum achievable resolution of the scanner?</p>

<p>- Leigh</p>

 

[richardsperry]

<p>Leigh,</p>

<p>The German states that the 1560dpi is the maximum resolution of the V600. While theoretically it doesn't make any sense to scan at 6400 when max resolution is 1600, and 3200 should suffice.</p>

<p>Experimenting with actual scanning results with different results than his conclusions. The scans look different, and behave different with typical post processing, say Unsharp Mask at the very least.</p>

<p>I definitely do not think that a 4000 dollar scanner is needed for negative proofing and posting film images to the web. I know I am in the minority here in supporting cheap low cost scanners, but I am just going off what I see. A Coolscan 9000 scan, sent out, is going to cost me about 40 dollars per 6x6 negative; that equates to 4 scans to buy the V500 new for my self.</p>

<p>Honestly, the thought has crossed my mind to buy one just as back up for when they are discontinued, and this one breaks down, as a spare. For all you pros, who sell enough prints to justify a 9000 scan or buying a drum scanner, by all means do so.</p>

<p>But it is certainly not the only option for a hobbyist photographer.</p>

<p> </p>

[zack_zoll]

<p>Richard, Leigh, etc. :</p>

<p>I'm just guessing here, but comments about the limitations of a scanner's dpi might have less to do with sensor resolution, and more to do with lens/processor resolution. I don't know the engineering specs of each model so please feel free to disagree with my point, but not my math. I know these are arbitrary numbers :)</p>

<p>Let's say you have a very small black to white gradient, and the scanner records that as a line of ten distinct pixels at a given resolution. That gives you ten distinct 'objects' in your scan of a given size. If you sharpen the heck out of your image, each one of those ten objects will have a slight border. Let's assume worst-case scenario, and say that the lens or the processor can only see those ten colours. I know that no scanner in the world is that bad, but go with me here.</p>

<p>If we quadruple the resolution while keeping the lens and processor the same, each 'object' recorded by the scanner will be four pixels. However, the image will look identical, because the scanner can only see ten colours. The resolution jump would imply that we are now getting 20 colours (since there are twice as many pixels going from black to white), but the lens and/or processor limitations are still 'squishing' that into ten colours. If we sharpen the image, we'll still get ten distinct objects of the same size, although the border will now be slightly smaller, since the pixels are all smaller. We won't get any more actual information that the original scan though.</p>

<p>The only thing we might get is more grain pattern in the scan, as the scanner is looking for smaller and smaller objects of the same ten colours. In other words, scanning at a higher resolution than the lens and/or processor can support may actually degrade your image.</p>

<p>I suspect is that this is what various people are talking about when they say that 'scanner X only has half of the listed resolution.' Obviously you can right click it and see that it really has more resolution than the review says, but it might not contain as much detail as the resolution numbers would indicate. This is probably lose in the German to English translation.</p>

<p>My thinking, Richard, is that the reason you're getting a better scan at 'full resolution, partial size' is because you're essentially giving the processor a larger sample space. By setting it at 25% you're throwing out 3 out of 4 pixels, but scanning at full res allows it to see more pixels and make a better decision about which three to toss out. Sony claims that the compression on their AVCHD movies works this way, and that it results in better image quality.</p>

[leighb]

<p>Sorry,l Zack, but you're totally missing the point.</p>

<p>You're confusing post-processing with what the scanner does. They're two entirely different and unrelated subjects.</p>

<p>You can't "quadruple" the resolution, or double it, or whatever. It is what it is. You can reduce it if you wish to get a smaller file size, but you can't increase it.</p>

<p>- Leigh</p>

 

[david_goldenberg1]

<blockquote>

<p>I definitely do not think that a 4000 dollar scanner is needed for negative proofing and posting film images to the web. I know I am in the minority here in supporting cheap low cost scanners, but I am just going off what I see.</p>

</blockquote>

<p>I absolutely agree! But, there really is a difference between a flat-bed scanner and a film scanner. I mentioned before that I had performed slanted-edge MTF tests on an Epson V500 and a Coolscan 8000. Here are graphs from those measurements. <br>

For those who aren't used to graphs of this sort, the vertical axis represents the fraction of the orginal contrast that is recovered in the scan, and the horizontal axis represents different spatial frequencies in cycles per inch. For the Epson, the maximum frequency is 3200 cycles per inch (one half of the maximum number of pixels per inch), while the maximum for the Coolscan is 2000 cpi. The individual curves are from scans made at different resolution settings.<br>

For the Epson, the curves with scan resolutions of 6400, 3200 and 2400 ppi are essentially identical, meaning that no additional information is gained by using a setting greater than 2400. At 1200 ppi, the Nyquist limit (600 cpi) is reached before the optical limit is. The relationship between MTF and resolution is somewhat arbitrary, but one interpretation is to say that the maximum frequency at which the MTF is greater than 10% corresponds to the maximum resolution. In this case, this occurs at about 700-800 cpi, which would correspond to about 1500 dpi. This is pretty close to what www.filmscanner.info says for the V600, using a different method (a resolution chart). <br>

For the Coolscan set for 4000 ppi, the MTF curve reaches 10% at about 1400 cpi, or 2800 dpi. Reducing the scan resolution to 3000 dpi causes a noticeable loss in information, and the MTF curve just reaches the Nyquist limit. With a 2000 dpi scan, the result is clearly limited by the scan resolution.<br>

For my slanted edge, I used a double-edged razor blade in a glass slide mount. This may not be ideal, since the blade is very reflective. But, I think that the comparison makes the relative performance of the two scanners pretty clear.<br>

Whether or not this difference is important, depends entirely on the user and application. In my experience, one of the practical differences is that scans from the Coolscan require minimum sharpening, with correspondingly less artifacts.<br>

As always, YMMV.<br>

David<br>

P.S. For more about MTF and the slanted edge test, see this page:<br>

http://www.imatest.com/docs/sharpness.html</p>

<p> </p><div></div>

[david_goldenberg1]

<p>Here is the graph for the Coolscan 8000</p><div></div>

[richardsperry]

David,

 

Do you have any scans, that can be posted, of the same negative from both scanners for comparison?

[david_goldenberg1]

<p>Richard,<br>

Funny you should ask! 2 1/2 years ago I started what turned out to be a very long thread, very similar to this one, by posting scans from the V500 and a Coolscan V:<br>

http://www.photo.net/medium-format-photography-forum/00RS5y?start=0<br>

(Depending on your browser you may have to go to the fourth page of the thread to actually see the scans. I had some file format problems.)<br>

There are lots of other comparisons like this on the web, and one can pixel peep to justify just about any conclusion one wants! As demonstrated by the this and other threads.<br>

David</p>

[richardsperry]

Thank you David for the link.

[zack_zoll]

<blockquote>

<p ><a name="00Z0q5"></a><a href="../photodb/user?user_id=6063681">Leigh B.</a> <a href="../member-status-icons"><img title="Subscriber" src="../v3graphics/member-status-icons/sub2.gif" alt="" /><img title="Frequent poster" src="../v3graphics/member-status-icons/1roll.gif" alt="" /></a>, Jul 08, 2011; 08:46 p.m.</p>

</blockquote>

 

<blockquote>

<p>Sorry,l Zack, but you're totally missing the point.<br>

You're confusing post-processing with what the scanner does. They're two entirely different and unrelated subjects.<br>

You can't "quadruple" the resolution, or double it, or whatever. It is what it is. You can reduce it if you wish to get a smaller file size, but you can't increase it.<br>

- Leigh</p>

</blockquote>

<p>Leigh, I think you may be misreading my post. With the exception of sharpening, I didn't discuss post-processing. The scanner itself processes the images. When I was discussing processors, I was referring to the one built into the scanner. If the processor built into the scanner is only able to process X amount of tones, then it will continue to see a maximum of that number regardless of how much resolution you give it to use as a sample size.</p>

<p>As an anology, take the Nikon D40 and similar cameras. My experience is that those cameras are much more likely to blow out highlight detail than the larger Nikons, even shooting in RAW. The sensor (the camera's eyes) may be the same as in a more expensive camera, but the processor (the camera's brain) in unable to comprehend the entire gamut of tones. Thus, highlights and shadows are occassionally clipped, even though the camera technically 'sees' the same thing, and you're theoretically not processing the image. To take the anology a step further, I have a friend who is legally unable to drive. He has better than 20/20 vision, and has no felony history. The reason he can't drive is that for some reason his brain (his processor) can't take the images from his eyes and create depth. Even though his eyes are great, he has no depth perception whatsoever, and can only guess at distances based on relative sizes. </p>

<p>I'm not saying that all, or any, flatbed scanners are this limited. I am, however, saying that the ability of the scanner to actually parse the information it is seeing is just as important as what it sees.</p>

 

<blockquote>

<p>You can't "quadruple" the resolution, or double it, or whatever. It is what it is. You can reduce it if you wish to get a smaller file size, but you can't increase it.</p>

</blockquote>

<p>Again, I wasn't referring to quadrupling the resolution in post. I was referring to scanning the image at a resolution that is four times higher. If the processor is maxed out at a given scanning resolution, scanning at a higher resolution will give the same or only slightly better results.</p>