Scanner for 6x7 color negatives

[AlanKlein]

<blockquote>

<p>"One scan can pick up the full range. More scans won't help what you capture in D/R or dMax."<br /> Not true either. If the exposure - brightness of scanning light source or acquisition time - is increased between consecutive scans, then obviously a greater density can be penetrated and hence a greater dynamic range can be captured.</p>

</blockquote>

<p>This sentence is the only thing in your post that responds to my point that the full range is captured with one scan.</p>

<p>More scans would only have a difference if the light source changes which is not available on scanners. Even if they are, or you can override the normal operation of the scanner, you would have to increase the light source to penetrate further into shadow areas. I assume it's in the manufacturer's interest to have calibrated the scanner to provide the brightest light that won't overload the CCD sensors to arrive at the highest dMax value to that particular machine. So, even if you were able to increase the light source, you would over drive the CCD's. </p>

<p>It is possible that the machine could combine two scans so that the over light scan only used the shadow areas and the normal scan uses the highlighted areas. Is that what you're arguing? If so, how is that done? Some sort of HDR?</p>

<p>What I'm more interested in knowing because of my own use, is whether I should scan flat with no adjustments in the scanner and make all edits in post. Or, adjust the white and black points on the histogram to points just outside the histogram range. I've been told I would get more data. But I'm not convinced because it seems that all that's happening in the latter procedure is that the points are being applied after the scan to the same data that would otherwise be available to the flat scan file. What's you take on this?</p>

 

[barry_r]

<p>"What I'm more interested in knowing because of my own use, is whether I should scan flat with no adjustments in the scanner and make all edits in post. Or, adjust the white and black points on the histogram to points just outside the histogram range. I've been told I would get more data. But I'm not convinced because it seems that all that's happening in the latter procedure is that the points are being applied after the scan to the same data that would otherwise be available to the flat scan file."</p>

<p>Alan: I'm going by memory here, but during my working days I used an Epson V750 and 11000XL, mainly running the Epson Scan program, for R&D not photographic applications. I don't remember where the actual setting is made, but by setting the histogram min and max grey levels, you end up with the 255 scanner levels (for 8-bit scans) between those two settings, not the default 0 and 255. This is better than scanning with the default and applying those settings afterward. If you compare the histograms using these two methods, the latter would have gaps since the scanned data is just being stretched to fit the 0 and 255 limits.</p>

 

[ondebanks]

<blockquote>

 

<p>Multi sampling and multi scanning give less sharp details because of combining samples so i understand from others who have tried them.</p>

 

</blockquote>

 

<p>Alan, as I said, multi scanning can misregister the iterated scans and hence blur details - but multi sampling emphatically <em>does not</em> blur details (and cannot, when you think about what is actually happening...I thought I had explained it rather clearly).</p>

 

<blockquote>

 

<p>There are others who claim that these techniques bring out the shadows better but I have yet to see a posted example that proves it.</p>

 

</blockquote>

 

<p>I ran such tests about 10 years ago, before I bought Vuescan. Here is a crop of a dark slidefilm photo [a short startrail image].<br>

First: normal sampling:</p>

 

 

<p><img src="http://imageshack.com/a/img633/9431/SriWRv.jpg" alt="" /></p>

<p>Now, multipass sampling:</p>

 

 

<p><img src="http://imageshack.com/a/img633/6209/nMe1Ik.jpg" alt="" /><br>

The improvement is obvious. Lower noise, smoother background sky, real signal reinforced, more faint stars revealed. Even the dirt smudges on the film are clearer!</p>

 

<blockquote>

 

<p>The dMax of the machine is set by the manufacturer to give the max light to penetrate on the first scan. Subsequent scans or samplings won't penetrate deeper. The light isn't a shovel. These are just selling points contrived by Vuescan.</p>

 

</blockquote>

 

<p>You are ignoring the nature of noise and measurement - and in the process, you are doing Vuescan (and its creator, Ed Hamrick) a grave injustice. Yes, you can reach into the shadows in a single sample. But you'll only do so at a poor signal-to-noise ratio. Multiple independent samples, when combined internally by Vuescan or any similarly-functional software, DO genuinely improve signal to noise, and hence pull more details (not resolution, but tonal differences) out of the shadow areas.</p>

 

[AlanKlein]

<p>The sampling techniques you show are blurring the background, something you can similarly do with"blur" and other tools in Photoshop. Showing additional star "spots" in the black background of space is not a real world test. I could do that changing the black point and white point level sliders. How about showing some real pictures that have a full range of details, DR and light so we can see the shadows before and after? The ones I've seen presented by others in years past using so-called special scanning software can be more simply done using the shadow slider in PS to bring out the details although with a increase in noise. </p>

<p>Additionally, you have not stated how multipass sampling can bring out more details. You're not digging a hole where each additional scoop gets you more dirt. The scanner's light when at its normal maximum setting, will provide the most light through the shadows. Doing it twice or three times won't penetrate the negative further. The dMax will not increase with more scans of the same negative. Assuming you can increase the scanner's light output, the amplifiers and sensors measuring the photons will cause distortions and more noise. If you reduce the light output, well then you'll get even less details than normal setting scans. </p>

<p>The fact is the scanner is limited to the optimum setting of the scanner's light output and receptive ability of its sensors. The manufacturer has tuned the setting for optimum combination maximum light for the sensors and signal amplifiers provided. Any changes will only cause distortion and/or less details.</p>

<p>Someone suggested slowing the scan speed so the light stay longer so the sensors pick up the data better. That sounds possible, I suppose, but can you change the speed of the scanner motor mechanism? Which scanners can you do that on? (I use a V600).</p>

<p>Believe me. I'm open to getting more details out of shadows. I'd love to get that out of my film. But "show me the money". Where is the proof? Show me real world samples, not a picture of space with clipped blacks and clipped whites.</p>

[ondebanks]

<blockquote>

<p>The sampling techniques you show are blurring the background, something you can similarly do with"blur" and other tools in Photoshop.</p>

</blockquote>

<p>It's not blurring! I made that clear already. The background becomes smoother in multi-sampling purely because of improved signal to noise, separately achieved at each and every pixel without reference to their adjacent pixels.</p>

<p>Blurring - of the Photoshop kind - means applying a convolution kernel which redistributes signal and noise around adjacent pixels. That trades a gain - improved local smoothness - against a loss - a decrease in spatial resolution ie. a softer image. But overall signal to noise doesn't change, because it's just pushing around the same original signal and the same original noise. Because of this, the background would become smoother in convolution-based blurring, but the stars would become harder to see. On the contrary, in my example of multi-sampling, the stars have become easier to see.</p>

 

<blockquote>

<p>Showing additional star "spots" in the black background of space is not a real world test.<br>

Show me real world samples, not a picture of space with clipped blacks and clipped whites.</p>

</blockquote>

<p>It's not the black background of space - it's the grey background of the real sky over my garden. How is that not a real world test? It's not some simulation. It's real. Also, there are no clipped blacks, other than in the few small dirt smudges. The sky background airglow - by definition present at all points in the image, including where the stars are - sets the absolute floor of intensity for an image like this. And as everyone here can see, the sky is rendered as shades of grey, not clipped black.</p>

<p>Clipped whites are not of relevance here: I clipped them post-scan so that the background improvements would be more visible - in order to provide you with the evidence you asked for.</p>

<blockquote>

<p>can be more simply done using the shadow slider in PS to bring out the details although with a increase in noise.</p>

</blockquote>

<p>Bingo. Anyone can shift the levels of an imge to an arbitary output intensity, but as you say, the boosted signal comes with boosted noise: there's no improvement in signal to noise. But there is an improvement in signal to noise with multi-sampling, and I've demonstrated it above.</p>

 

<blockquote>

<p>Additionally, you have not stated how multipass sampling can bring out more details. You're not digging a hole where each additional scoop gets you more dirt.</p>

</blockquote>

<p>It's not a great analogy, but I'll play with it: You are re-digging the same hole, over and over. The goal is to dig a hole of the correct size - analagous to measuring a pixel at the correct intensity. Because of digging imprecision, each hole you dig will be a little too big, or too small, or maybe almost just right now and again. The point is that the <em>average</em> size of all the holes will be closer to the correct size than any individual hole attempt...the <em>average</em> intensity of all the samples at that pixel location will be closer to the correct intensity than any individual sample (bar the occasional statistically lucky one).</p>

<p>This is how and why multi-sampling works. In the holes analogy, the error in the hole size is the noise in the measurement. Taking the average reduces that error; reduces that noise. When you take N independent samples of something which follows a Poisson distribution (as photon counts of light do) or a Gaussian distribution (as scanner readout noise does), you improve the signal to noise by a factor equal to the square-root of N. 16 samples (my default setting in Vuescan) therefore improves S/N by a factor of 4. It's a little more complicated by the fact that both types of noise distribution are present at a scanned pixel to different degrees, but as I said, a standard noise model (combining the noise contributions in quadrature) accounts for that.</p>

<p>The next thing to consider is this: if 16-pass sampling improves the S/N of all scanned pixels, from the darkest to just below the saturation point, by a factor of sqrt(16)=4, does that mean that all pixels <em>look</em> 4 times better? The answer is 'no', because our human vision cannot discriminate as well between high S/N information as it can between low S/N information. An image detail with S/N of 100 looks virtually the same as one with S/N of 400. But a detail with S/N of 1 looks horrendous compared to one with S/N of 4. In fact, S/N of 1 is at the threshold of discrimination from noise (it would satisfy an engineer's definition of a detectable signal, but not an astronomer's), while S/N of 4 is clearly visible; which is why more stars emerged from the previously obscuring noise when I activated multi-pass scanning.</p>

<p>That is why I chose the example of a very dark (but not black) sky with faint stars for my test comparison. The relative gains in signal to noise are indeed the same for the brighter stars, but less perceptible, so I had no qualms about clipping them.</p>

<blockquote>

<p>Assuming you can increase the scanner's light output, the amplifiers and sensors measuring the photons will cause distortions and more noise.</p>

</blockquote>

<p>If you send more light through the film to the sensor, noise impact will decrease, not increase - this is the nature of Poisson statistics. But you will saturate the pixels which receive light through clearer areas of the film; you won't capture that end of the photo's DR. So it's not a free lunch.<br>

I'm not sure what you mean by "distortions" in this context?</p>

[AlanKlein]

What is changed in the scanner during multipass?

Averaging the results doesn't provide deeper

penetration. You can't get more than the dMax

maximum of the scanner's design.

 

In any case, stars on a gray background is not a real

world test of what most photographers shoot. Show

me what you can do with a scene that has a person,

trees, sky, and shadows and highlights. Provide files

of a flat scan, auto scan using Epsonscan,

scan using Vuescan and a flat scan edited by

Photoshop. Then we all can get an idea of the

differences in results of each method.

[ondebanks]

<blockquote>

<p>What is changed in the scanner during multipass?</p>

</blockquote>

<p>The amount of intensity information it acquires for each pixel location.</p>

<blockquote>

<p>Averaging the results doesn't provide deeper penetration.</p>

</blockquote>

<p>It does, if a single sample has not recorded the densest areas with sufficient S/N.<br>

Observational astronomy research (which is my profession, BTW) would still be in the 19th Century if this were not the case!</p>

<blockquote>

<p>You can't get more than the dMax maximum of the scanner's design.</p>

</blockquote>

<p>You can, because the dMax is specified from the performance of a single pass. Mainly because most scanners don't support multipass so it maintains a level playing field, and perhaps also to prevent manufacturers from advertising unrealistically good numbers that could only be achieved with a zillion passes.<br>

Also bear in mind the <a href="http://www.kenrockwell.com/tech/scantek.htm">caveat on whether a scanner manufacturer is quoting dMax (possibly attained only by overexposing the highlights) or DR</a>.</p>

<p>In the same way, digital (still) camera manufacturers quote DR based on a single frame; because again, averaging frames can increase DR at the faint end, but few photographers combine frames, and there has to be a common standard.</p>

<p>In the digital video camera sector, because taking rapid sequential frames is the nature of the camera, the opposite seems to be true: firms quote enormous DR for their cameras, as they can take advantage of temporal frame averaging.</p>

<p>Look, this is all basic measurement science. I don't get why you are still so opposed to what has long been established and proven fact. If you are unfamiliar with it, fine; hit google or wikipedia with some of the terms I used above, and you soon will understand it. </p>

<p>I'm not going to the trouble of doing fresh scans just to prove this all over again. If you still need convincing, get the free trial of Vuescan Pro Edition (it just watermarks the scans) and see for yourself on a dark Velvia slide.</p>