OK, my reply was based on my Konica-Minolta driver, which separates out physical (at scan) per-channel exposure adjustments from (post-scan) levels and curves. I do also have a V750, so I will do an experiment and report back, although a longer scan time may only prove that the driver is doing something - whether that's adjusting exposure time, boosting signal gain or simply remapping tone values, who knows. Using Silverfast's curves and exposure adjustments, the Epson can really pull a lot of detail out of thin negs, but I assumed it was through remapping and interpolation.
Scanning Black and White Film as Negative or Positive?
-
H
- Thread starter brian steinberger
- Start date
Recent Classifieds
-
For Sale FS: Linhof Recessed lens board
- Started by dirkfletcher
-
For Sale Leica M4-2
- Started by chriscrawfordphoto
-
For Sale FS: Toyo Sliding Back for 4x5 Graflok
- Started by B.S.Kumar
Forum statistics
Well, I've made some (unscientific) test scans of dense and thin materials, with and without adjustments. I've timed each scan, and adjusted the linear (default) scans in PS to match those optimised pre-scan, and then compared the images for noise in each channel. And I've found nothing to challenge my assumption that the Epson V750 is a dumb digital capture device with a smart back end. No appreciable difference in scan times between dense and thin originals, or between optimised and default scans. No appreciable difference in image noise between scans optimised in Silverfast and in PS.
For what it's worth, I was scanning with Silverfast at 16-bit colour positive (colour negative materials).
For what it's worth, I was scanning with Silverfast at 16-bit colour positive (colour negative materials).
Hi
I've never used silverfast (aside from trying it at first) so I can't comment on that product. However using Epson scan I do get time changes.
Getting back to addressing the other issue:
Scanning in Neg mode on the Epson software (as I said earlier) makes the scanner work differently and drives the capture of data differently. Essentially its now NON LINEAR. So any adjustments of levels (black and white points) introduce curves as can be seen in this screen grab
scan-fig1 by aquinas_56, on Flickr
and also this screen grab
scan-fig2 by aquinas_56, on Flickr
so if you have a grasp of integer maths you'll know that you're getting different numbers which are unrecoverable in the same manner by "undoing" that curve.
so as I said in the start, you do get different things.
PS: why don't you (the reader) WITH YOUR OWN SCANNER attempt to verify or explore this issue your self.
I've put up a quick blog post with some more examples. If anyone has any information to add to this please by all means post it. By all means examine this information and see if it can be repeated at your end. Or just keep doing what ever your doing ... I don't really mind
I've never used silverfast (aside from trying it at first) so I can't comment on that product. However using Epson scan I do get time changes.
Getting back to addressing the other issue:
Scanning in Neg mode on the Epson software (as I said earlier) makes the scanner work differently and drives the capture of data differently. Essentially its now NON LINEAR. So any adjustments of levels (black and white points) introduce curves as can be seen in this screen grab
scan-fig1 by aquinas_56, on Flickr
and also this screen grab
scan-fig2 by aquinas_56, on Flickr
so if you have a grasp of integer maths you'll know that you're getting different numbers which are unrecoverable in the same manner by "undoing" that curve.
so as I said in the start, you do get different things.
PS: why don't you (the reader) WITH YOUR OWN SCANNER attempt to verify or explore this issue your self.
I've put up a quick blog post with some more examples. If anyone has any information to add to this please by all means post it. By all means examine this information and see if it can be repeated at your end. Or just keep doing what ever your doing ... I don't really mind
Last edited by a moderator:
I can scan flat and make the same kind of change in PP using PS Elements. The scanner software is doing the same thing after the scan except it seems like it's changing it during the scan. But it isn't as far as I can tell. How could it other than by changing speeds. Then you would have to do two scans (something that Silverfast does if set to do). Then combine afterward like HDR.
The one area where I can agree the scanner may make a change is if by moving the black and white points, the data file will have more data of actual exposed light on the film than if you captures the light 0-255 from the whole film. For example, if the range of a particular picture was 20-230 and you moved the points in so the full capture was from 20-230 rather than 0-255, than it is possible that you'd have more data to work from. Of course, the physical scan is the same either way. But the post processing by the scanner program provide more data about the actual light that makes the picture. Does anyone know if this actually happens?
The one area where I can agree the scanner may make a change is if by moving the black and white points, the data file will have more data of actual exposed light on the film than if you captures the light 0-255 from the whole film. For example, if the range of a particular picture was 20-230 and you moved the points in so the full capture was from 20-230 rather than 0-255, than it is possible that you'd have more data to work from. Of course, the physical scan is the same either way. But the post processing by the scanner program provide more data about the actual light that makes the picture. Does anyone know if this actually happens?
are you saying scan flat (meaning the curve) with the scanner software in negative mode?
if so, then no you can't (or at least if you can i'd like you to demonstrate that rather than just say you can)
you seem quite thin on backing up anything you say and quite thick on saying no it isn't
so first you say it is, but then admit that it seems like it isn't ... interesting
as I said initially, you're thinking like this is a digital camera. This is scanning a negative, unlike a scene this is actually all there is to it.
as far as I know the only way to do that is to tweak your photomultiplier which is what drum scanners do. So from the absolute theoretical standpoint (and ignoring all the reality inbetween) yes, your right.
For clarity this is the typical scan range of a 4x5 black and white neg, including base fog and shadow cast by the film holder
as you can see, the darkest parts and thinnest parts fall in the middle 2/3 of the scanners range. Not bad by design actually as this means you avoid "noise" areas inherent in cheaper circuits.
So as mentioned you don't get at more shadow detail here by changing the lower end slider of the Epson because there isn't much to be had down there because there isn't any more density in the film. This is true except for colour where it has more density down there and as has been identified we get longer scan times ... which effects the amount of data that the scanner pulls out of the dense areas.
tell me, have you actually examined any of this or is it all just "thought experiment"??
I know this happens in drum scanners and I know that this does not happen in Epson flatbeds or Nikon coolscans (although you can tweak analog gain there to some extent).
I still feel that you're a) not reading anything I put forward, b) not seeing my original point c) not comprehending the importance of curve operations on integers in the negative scan mode.
But aside from the small losses of information due to the data transformations in the curves, I agree that you do not get access to a better range of data in a black and white scan by scanning altering the scan. That you are unable to the significance of what happens when you unsquish that curve forced upon you by choosing "negative mode" is not worth discussing further.
Last edited by a moderator:
Can you clarify your last paragraph? The first sentence says thats you do not get "a better range data" by altering the scan. The second says "unsquishing the curve" is "significant". If you don't get "a better range data", how can "unsquishing the curve" be "significant"?
Firstly, Pellicle, I have looked at your blog and find it really interesting, thanks.
Alan, if you set black and white points in the scanning software, the software will expand the data to fill the alloted range. If you choose 16-bit, it should be able to do this with few or no banding artifacts, depending on how much expansion is required. This is how I understand the process, and I've yet to be persuaded otherwise! Likewise, a very thin scan (ie one with a squeezed range of tones) can be 'expanded' very successfully in PS, as long as it is 16-bit.
I have never found an advantage in doing it one way or another, but doing it at the scanning stage just feels right to me. The Silverfast preview and densitometer are good enough to give me a pretty good idea of what I'm going to achieve.
Personally, I always scan negatives as positives, setting black and white points manually for each channel if required. Once inverted in PS, I have a reasonably good springboard for tweaking. I never use NegaFix or the other proprietary negative presets as I find they often clip too harshly. I like a bit of headroom in my scans.
Alan, if you set black and white points in the scanning software, the software will expand the data to fill the alloted range. If you choose 16-bit, it should be able to do this with few or no banding artifacts, depending on how much expansion is required. This is how I understand the process, and I've yet to be persuaded otherwise! Likewise, a very thin scan (ie one with a squeezed range of tones) can be 'expanded' very successfully in PS, as long as it is 16-bit.
I have never found an advantage in doing it one way or another, but doing it at the scanning stage just feels right to me. The Silverfast preview and densitometer are good enough to give me a pretty good idea of what I'm going to achieve.
Personally, I always scan negatives as positives, setting black and white points manually for each channel if required. Once inverted in PS, I have a reasonably good springboard for tweaking. I never use NegaFix or the other proprietary negative presets as I find they often clip too harshly. I like a bit of headroom in my scans.
Hi
none the less let me have a go at this. Lets pretend that there are two types of operations you can do on your scanner data (represented graphically for the user of the computer):
- a stratight line operation
- a curved line operation
now a straight line operation is simple division by a linear operator, the changes to each column of the bar graph which represents the data in the histogram is a smooth and incremental change compared to its neighbor.
still not easy to undo, but it does represent close to the original information captured by your scanners sensor
now, lets consider a curve. The way you get a curve is to apply a more complex transform to the data, such as a log() function or some more complex variation of that (such as when you put additional points in the curve).
so lets use a function like COS() start with an integer like 36 and perform a cos on it, you get 0.80901699437494742410229341718282 ... all well and good. However if you can not represent that number in enough significance you have problems. Lets say you can only represent it as 0.809017 ... when you attempt to perform the inverse operation on that (I'm guessing you didn't read about orthogonal in that blog post) you will not get 36 back.
Now the problem for this comes from the fact of life that shadows in digital form get far smaller integers to represent them than do high lights. (A quick web search reveals a tutorial on this aspect here, fig below from that tutorial).
so the levels which you have to distinguish the shadow levels in the image are low to begin with, performing operations on them such as done by application of "curves" will result in the loss of information from what you have captured in a linear scan. Attempting to undertake reversal of this (with subsequent curves application) will result in some pixels which were slightly but distinctly different from their neighbor being moved to the same value, and perhaps others being made different which were previously the same.
does that explain what I said better?
ok, but first can you clarify how much maths you understand? Do you understand the difference between floating point maths and integer maths?
none the less let me have a go at this. Lets pretend that there are two types of operations you can do on your scanner data (represented graphically for the user of the computer):
- a stratight line operation
- a curved line operation
now a straight line operation is simple division by a linear operator, the changes to each column of the bar graph which represents the data in the histogram is a smooth and incremental change compared to its neighbor.
still not easy to undo, but it does represent close to the original information captured by your scanners sensor
now, lets consider a curve. The way you get a curve is to apply a more complex transform to the data, such as a log() function or some more complex variation of that (such as when you put additional points in the curve).
so lets use a function like COS() start with an integer like 36 and perform a cos on it, you get 0.80901699437494742410229341718282 ... all well and good. However if you can not represent that number in enough significance you have problems. Lets say you can only represent it as 0.809017 ... when you attempt to perform the inverse operation on that (I'm guessing you didn't read about orthogonal in that blog post) you will not get 36 back.
Now the problem for this comes from the fact of life that shadows in digital form get far smaller integers to represent them than do high lights. (A quick web search reveals a tutorial on this aspect here, fig below from that tutorial).
so the levels which you have to distinguish the shadow levels in the image are low to begin with, performing operations on them such as done by application of "curves" will result in the loss of information from what you have captured in a linear scan. Attempting to undertake reversal of this (with subsequent curves application) will result in some pixels which were slightly but distinctly different from their neighbor being moved to the same value, and perhaps others being made different which were previously the same.
does that explain what I said better?
Pellicle, it seems to me your explanation is less relevant for 16-bit data.
OK I think we're finally all on the same page. So let's expand that page. Leaving aside 16 vs. 8 bits for the moment, does setting the white and black points to a scanned image with a range of let's say 20 -230 to output at 0-255 create a file with more data? If so, then it makes sense to set the the points before the scan. If it doesn't than you can wait to do the adjusting after the scan in post.
Now let's talk about 16 bits. Since you are dividing up the data in to so many more bits, does it matter with 16 bits to adjust the black and white points before? Even assuming that this works with 8 bits as describe above.
Now let's talk about 16 bits. Since you are dividing up the data in to so many more bits, does it matter with 16 bits to adjust the black and white points before? Even assuming that this works with 8 bits as describe above.
Testing
OK I decided to look at the data I got on scans I did a few months ago and see if this would shed light on this. I scanned the same photo with and without corrections during the scan. Film was Velvia 50 chrome. 6x7 medium format. 2400 resolution. 24 bit (8 bit per channel). Epson V600 scanner. Epson scanner software. Both scans had some of the black film holder borders around the photo inside the crop marquis. Exactly same crop on both scans.
Scan 1. Scanned flat. No adjustments during scan. ICE off. TIFF File: 109.22MB. RGB Histogram indicates 38,173,792 pixels. Range 0-185.
Scan 2. Scanned with auto adjustments set in scanner. Color Control Continuous Auto Adjustment (in Configuration) 2.2 gamma. No other exposure or color adjustments set. ICE was set however. TIFF File: 109.12MB Histogram indicates 38,136,780 pixels. Range 5-255.
Results: Scan 2 with auto giving 5-255 was very slightly smaller than the scanner output file in Scan 1. There's only 37,000 pixels or .01 percent difference. That could be because the ICE was turned on in Scan 2. Regardless, that's an insignificant amount a data and it's less data in Scan 2 with adjustments than in Scan 1 without adjustments. Moving the white point before the scan from 255 to 185 using Auto scan adjustments gave no additional data.
THERE ARE NO DIFFERENCES IN QUANTITY OF CAPTURED DATA BETWEEN SCAN 1 AND SCAN 2 REGARDLESS IF YOU SET BLACK AND WHITE POINTS DURING SCAN OR NOT. It appears that the scanner is just applying a setting to the data captured by the scanner circuits. You could take the scanner's flat scan file and adjust in Post and get the same results. Of course, the advantage of doing it Post is that no subsequent scans would be required. Just adjust the settings in Post. If your settings are wrong during the scan, then you have to spend time re-scanning.
Interestingly, when I adjusted Scan 1 afterwards in Elements, I got the same pixels that came out of the scanner although the file size was slightly larger. The total pixels when adjusted in Elements are the same as the test where the adjustments for points were made in the scanner. Adjusted levels in Elements using Auto adjust. Saved TIFF File: 109.92MB. Range 0-255. Histogram indicates 38,173,792 pixels, same as what came out of the scanner. That's good to know because it shows that Elements is not losing any pixels when it auto adjusts Levels.
I didn't do any 16 bit scans (48 bits). However, I would think the results would be similar because you'd be capturing even more data at each of the 255 points.
Here's the final picture I posted in Flickr. Hubbard Glacier | Flickr - Photo Sharing!
OK I decided to look at the data I got on scans I did a few months ago and see if this would shed light on this. I scanned the same photo with and without corrections during the scan. Film was Velvia 50 chrome. 6x7 medium format. 2400 resolution. 24 bit (8 bit per channel). Epson V600 scanner. Epson scanner software. Both scans had some of the black film holder borders around the photo inside the crop marquis. Exactly same crop on both scans.
Scan 1. Scanned flat. No adjustments during scan. ICE off. TIFF File: 109.22MB. RGB Histogram indicates 38,173,792 pixels. Range 0-185.
Scan 2. Scanned with auto adjustments set in scanner. Color Control Continuous Auto Adjustment (in Configuration) 2.2 gamma. No other exposure or color adjustments set. ICE was set however. TIFF File: 109.12MB Histogram indicates 38,136,780 pixels. Range 5-255.
Results: Scan 2 with auto giving 5-255 was very slightly smaller than the scanner output file in Scan 1. There's only 37,000 pixels or .01 percent difference. That could be because the ICE was turned on in Scan 2. Regardless, that's an insignificant amount a data and it's less data in Scan 2 with adjustments than in Scan 1 without adjustments. Moving the white point before the scan from 255 to 185 using Auto scan adjustments gave no additional data.
THERE ARE NO DIFFERENCES IN QUANTITY OF CAPTURED DATA BETWEEN SCAN 1 AND SCAN 2 REGARDLESS IF YOU SET BLACK AND WHITE POINTS DURING SCAN OR NOT. It appears that the scanner is just applying a setting to the data captured by the scanner circuits. You could take the scanner's flat scan file and adjust in Post and get the same results. Of course, the advantage of doing it Post is that no subsequent scans would be required. Just adjust the settings in Post. If your settings are wrong during the scan, then you have to spend time re-scanning.
Interestingly, when I adjusted Scan 1 afterwards in Elements, I got the same pixels that came out of the scanner although the file size was slightly larger. The total pixels when adjusted in Elements are the same as the test where the adjustments for points were made in the scanner. Adjusted levels in Elements using Auto adjust. Saved TIFF File: 109.92MB. Range 0-255. Histogram indicates 38,173,792 pixels, same as what came out of the scanner. That's good to know because it shows that Elements is not losing any pixels when it auto adjusts Levels.
I didn't do any 16 bit scans (48 bits). However, I would think the results would be similar because you'd be capturing even more data at each of the 255 points.
Here's the final picture I posted in Flickr. Hubbard Glacier | Flickr - Photo Sharing!
Last edited by a moderator:
Nice pic. Glad you've found your method at last. As I say, I've never found any advantage to doing it one way or the other, but I do like doing the adjustments in the driver (with headroom at each end to allow for imprecise histograms), simply because I like having a nice, full range of tones to start off with in PS.
I think your method is fine. Your rationale makes sense. I like the idea of headroom. Most of the pictures I scanned that are on Flickr used the scanner's Auto adjust. There was subsequent adjustments in Elements. But that doesn't provide any real headroom.
The thing that my little review did though was convince me that all this hype about the scanner getting better scans with all these adjustments are just hype. The programs seem to be just post editing like any other editor.
The thing that my little review did though was convince me that all this hype about the scanner getting better scans with all these adjustments are just hype. The programs seem to be just post editing like any other editor.
- Joined
- Jul 4, 2009
- Messages
- 146
- Format
- 35mm
Is this a troll? Seriously, you are measuring the number of pixels??
I think you can try the same experiment one more time. This time chose 1 bit black and white (not grayscale mode). I think you will be shocked to find that this also captures the same number of pixels therefore proving that it contains the same amount of image data. Feel free to back this up by adjusting your 1 bit image in Photoshop until it looks like the other two images. Please post your results.
Yes, he's confusing image data with image size, but the important thing is he's satisfied with his results, and I think, muddled though some of the thinking may be, he's found the solution to his problem (is there a difference between setting black and white points in scanning software or post-processing) - something that none of us has been able to supply or at least articulate.
I don't understand the 1 pixel suggestion. You'll have to explain. Sorry if I'm not as expert as you. But my Histograms look the same whether I set black and white points in the scanner or in Elements. So explain to me why it should be different and how. I really am open to learning. I just have not found anyone who has proven anything to me about this issue. How would I see the difference and measure it?
I'm sure that 16 bits will provide more precision on this. At the end of the day if you don't see any differences and you don't prefer to work this way then by all means don't. Personally I just always find that working this way gets more satisfactory results (more so with colour negative than black and white where the useful data range is further away from the floor noise of the scanner).
I'm not a salesman on this and so I really don't care what people do. Nor am I being paid to be an educator to design tutorials for people who seem to be requiring to be let by the resisting hand. I've provided quite a bit of start and evidence to support why I prefer to work this way to allow anyone with a scanner and some interest to follow up.
Allan If you have a scanner and photoshop you now have enough information here to answer your questions. I'm happy to assist in learning to fish, but I don't give out fish for free. Jesus did that and people are still following him.
2 minutes well spent
[video=youtube;qGFjrsfaatY]http://www.youtube.com/watch?v=qGFjrsfaatY[/video]
I'm willing to bet they don't ... they may look similar but I'm willing to bet they don't look "the same"
ok, I'm out on this one folks
If my analysis was wrong, please provide your analysis. I would like to see how doing black and white point adjustments before the scan provides more data to use in Post rather than scanning with no adjustments. Frankly, it makes sense to me as a possibility. I did my analysis trying to see what the results would be. If my analysis was faulty, OK, no problem. I'm open to learning but some of you calling me names or leaving the conversation doesn't prove anything. Show me how you get more data and can measure it.
- Joined
- Jul 4, 2009
- Messages
- 146
- Format
- 35mm
I am very sorry if I sounded harsh. I have some wonderful negative scans from my Coolscan from when I just used the Nikon software "negative" scan mode. I only started looking into this stuff when I was stuck for a time with a much inferior scanner that did not give good results "out of the box" and caused me to explore how I could maximize the results from it. Apart from that was my frustrations with the occasional frame that would confuse the software leaving me sometimes with images that had opposite colour casts between the shadows and the highlights. These were frustrating and hard to work with. My ultimate goal is an objective method for negative scans that gives a nice flat image for Lightroom with the maximum image data. I may throw some of that data out artistically in the processing but I don't want the image to start out that way. I want those decisions to be mine.
How much this tweaking matters really depends on your requirements and the capability of the hardware. If the hardware is capable it may still give good results even if you are not squeezing every last drop out of it so-to-speak. If it will not be mangled much in processing then even eight bits can be fine. Especially if you start out with a clean eight bit image. You can have more megapixels and less detail as we all know.
My goal in all of this is not to get people to work a certain way but just to convey a better understanding of how it all works. You don't need to understand how the engine works to be a good driver but it does help when the car refuses to do what you want it to.
How much this tweaking matters really depends on your requirements and the capability of the hardware. If the hardware is capable it may still give good results even if you are not squeezing every last drop out of it so-to-speak. If it will not be mangled much in processing then even eight bits can be fine. Especially if you start out with a clean eight bit image. You can have more megapixels and less detail as we all know.
My goal in all of this is not to get people to work a certain way but just to convey a better understanding of how it all works. You don't need to understand how the engine works to be a good driver but it does help when the car refuses to do what you want it to.
Thanks for your response. I appreciate the problems you've had with getting good scans. For the most part, I've been pretty lucky in using default settings and getting fairly decent scans from the Epson program for my V600 flat bed on both negative and chrome film in 120 and chrome film in 35mm. You can check these out using the link below. Of course I did post processing in Elements as well. What I'm trying to understand though is what is the difference in data captured by making adjustments before or after. I've read of so many people who spend enormous amounts of time tweaking the scan settings and getting frustrated. Are they wasting time that could be better spent in Post by scanning "flat"? Or am I not getting the best I can during the scan mode by not using the scanner adjustment settings?
So, I'm trying to understand how the scanner creates its image scan file. Let's see if I can ask the questions in simple ways so it's clear what I'm looking for:
1. If I have a film that has a histogram of 0-165, will I get more data in the scan file to work on if I move the white point in the scanner from 255 to 165 rather than leaving it "flat" and scanning the full 0-255?
2. How could I measure the difference between moving it and scanning "flat" assuming pixel count and file size does not provide that information as someone stated above?
Thanks. Alan
So, I'm trying to understand how the scanner creates its image scan file. Let's see if I can ask the questions in simple ways so it's clear what I'm looking for:
1. If I have a film that has a histogram of 0-165, will I get more data in the scan file to work on if I move the white point in the scanner from 255 to 165 rather than leaving it "flat" and scanning the full 0-255?
2. How could I measure the difference between moving it and scanning "flat" assuming pixel count and file size does not provide that information as someone stated above?
Thanks. Alan
A somewhat old thread, but I wanted to offer that while scanning negs as positives can be a powerful tool, I'd suggest not to use the Photoshop invert function to flip it. I avoid making any large tonal or color shifts in Photoshop, and especially using that invert function. It has to do w/ the math behind photoshop, and about it not actually being 16 bit, but instead 15+1 bit, which in short leads to it being pretty destructive to files w/ big moves. Scanning software is usually much better written for this function. Ditto for RAW processing, and why there's been a heavy trend of doing the big tone and color work during the raw processing stage, as it makes for a better file. For some reason this information isn't really passed around much though.
With that, scanning as a positive can be great, and if you're doing it I highly highly suggest the cfsystems colorperfect plug in for inverting positive scans. It's a bit raw of a UI that takes a while to learn, but it's a powerful plug in that essentially replaces what scanning software does w/ some better math build a nice file out of a raw scan. It lets you control everything, even how the highlights shoulder. It works best when the scanner is set up to do linear "gamma 1" scan (can be done w/ nikon scan), but it will work otherwise, as long as the file is 16bit.
With that, scanning as a positive can be great, and if you're doing it I highly highly suggest the cfsystems colorperfect plug in for inverting positive scans. It's a bit raw of a UI that takes a while to learn, but it's a powerful plug in that essentially replaces what scanning software does w/ some better math build a nice file out of a raw scan. It lets you control everything, even how the highlights shoulder. It works best when the scanner is set up to do linear "gamma 1" scan (can be done w/ nikon scan), but it will work otherwise, as long as the file is 16bit.
Last edited by a moderator:
Wow, I just stumbled onto this thread and read the whole thing. I have found great value in Pellicle's approach.
Alan, I think the fundamental difference between you and Pellicle is you are talking quantity of data, and Pellicle is talking quality of data.
You can't increase the amount of pixels you get with his approach, and he isn't suggesting that. He is suggesting that the pixels you do capture (the same number) will be qualitatively better - meaning that more of the pixels you get will accurately reflect image values, rather than noisy data. Pellicle is suggesting that this noise comes from two sources: the raw hardware noise at the edges of circuit performance, and algorithmic "noise" caused by pixel values being mathematically distorted during the process of capture.
Pellicle's workflow attempts to limit both, resulting in scans with overall higher quality. This higher quality gives you a better raw source of pixels to work with, it does not give you more pixels to work with.
You won't be able to see the difference by looking at pixel counts or by comparing histograms. You will only be able to see the difference by looking at the image itself following both workflows.
Pellicle's workflow requires more work, but allows for less automated mathematical imposition which cannot be reversed later in Photoshop. Pellicle claims (and I agree) that the algorithms applied are approximations that can't take into account all the qualitative differences in scenes, and as a result will cause you to lose some image precision in some scenes. He further claims that once you understand the workflow, the difference in effort is small, for a large qualitative gain in scan result (again, which can't be measured by comparing histograms or counting pixels).
I tend to agree with him on both points.
But your point that the automated scan of the negative is good enough is a perfectly reasonable conclusion for you. If you feel that the quality of the scan meets your needs, there is no need for you to do further work. And note, we are talking quality of pixel, not quantity of pixel.
Alan, I think the fundamental difference between you and Pellicle is you are talking quantity of data, and Pellicle is talking quality of data.
You can't increase the amount of pixels you get with his approach, and he isn't suggesting that. He is suggesting that the pixels you do capture (the same number) will be qualitatively better - meaning that more of the pixels you get will accurately reflect image values, rather than noisy data. Pellicle is suggesting that this noise comes from two sources: the raw hardware noise at the edges of circuit performance, and algorithmic "noise" caused by pixel values being mathematically distorted during the process of capture.
Pellicle's workflow attempts to limit both, resulting in scans with overall higher quality. This higher quality gives you a better raw source of pixels to work with, it does not give you more pixels to work with.
You won't be able to see the difference by looking at pixel counts or by comparing histograms. You will only be able to see the difference by looking at the image itself following both workflows.
Pellicle's workflow requires more work, but allows for less automated mathematical imposition which cannot be reversed later in Photoshop. Pellicle claims (and I agree) that the algorithms applied are approximations that can't take into account all the qualitative differences in scenes, and as a result will cause you to lose some image precision in some scenes. He further claims that once you understand the workflow, the difference in effort is small, for a large qualitative gain in scan result (again, which can't be measured by comparing histograms or counting pixels).
I tend to agree with him on both points.
But your point that the automated scan of the negative is good enough is a perfectly reasonable conclusion for you. If you feel that the quality of the scan meets your needs, there is no need for you to do further work. And note, we are talking quality of pixel, not quantity of pixel.
Chuck: Thanks for your thoughts. I've compared both my scans where one scan was scanned flat and was post processing in Auto Levels in PSE-8. It matched the image results in Auto set before the scan. So that test didn't seem to prove that one way is superior to the other. However, I might be fooled because I'm depending on my eyes in a small image on a computer screen.
Here are questions I asked previously but didn't get responses. Maybe you can answer them. I'm trying to understand how the scanner creates its image scan file. That may answer the question of which way is better.
1. If I have a film that has a histogram range of 0-165, will I get more data in the scan file to work on if I move the white point in the scanner from 255 to 165 rather than leaving it "flat" and scanning the full 0-255? In other words, will the data file at the end of the scan have more data representing the 0-165 then the same 0-165 area if I scanned "flat" from 0-255?
2. How could I measure the difference between moving it and scanning "flat" assuming pixel count and file size does not provide that information as someone stated above?
Thanks. Alan
Here are questions I asked previously but didn't get responses. Maybe you can answer them. I'm trying to understand how the scanner creates its image scan file. That may answer the question of which way is better.
1. If I have a film that has a histogram range of 0-165, will I get more data in the scan file to work on if I move the white point in the scanner from 255 to 165 rather than leaving it "flat" and scanning the full 0-255? In other words, will the data file at the end of the scan have more data representing the 0-165 then the same 0-165 area if I scanned "flat" from 0-255?
2. How could I measure the difference between moving it and scanning "flat" assuming pixel count and file size does not provide that information as someone stated above?
Thanks. Alan
Hi Alan,
I'll try to answer.
QUESTION 1
========
No, you will not get more data. In this answer I am assuming by "more" you mean it in the obvious sense: more pixels (or in computer jargon, more bytes) of data. Instead, the software will be able to map the 0-255 to finer changes in the raw measurements. In other words, a change of 1, say from 90 to 91 in value will represent a smaller change in tone.
This would be good, if the software didn't also apply a bias (or as Pellicle states, "change the 'curve'") at the same time, which is the fundamental issue Pellicle is raising. This bias means that the amount of change in tone between, say, value 90 and value 91 is not the same as the amount of change in tone between value 150 and 151, because the software is applying an additional function to the result spreading the tones in the range of 0-165 to the values 0-255. It is doing this non-linearly, or putting it in simple terms, applying a bias to the values.
You cannot change this back in Photoshop - this mathematical function cannot be reversed. Pellicle gets deeper into the reason for this, but it boils down to this: the scanner sees extremely fine changes in tone, which it maps to a (relatively speaking) very course set of fixed values (0-255 when using 8 bits). This process (in the software) is taking a floating point number between 0 and 1 from the CCD hardware (say, .25598337221) and mapping that to a fixed integer between 0 and 255 (say, 50). The next value from the CCD might be .25648825443, which might also get mapped to 50, unless a bias is being applied, in which case it could get mapped to 51.
Put yet another way, the scanner's software is automatically compressing or expanding certain value ranges outside your control when it automatically introduces a bias function.
Since Photoshop (or whatever editor you use) doesn't know it is getting values that have been biased by a function, you cannot easily (if at all) use the curves in Photoshop to counteract the bias that was done by the scanner. It just complicates things.
However, all that said, it is entirely possible that the best scan from a particular image might come from a scan of the negative, with a bias curve automatically applied. It all depends on the image. Converting from raw CCD data (which is basically analog), to a set of discrete digital values is not a process where one size fits all. That is the issue Pellicle and others have with relying on the automation in the scanner. In many cases, the math is not a good fit for the data, and the scan suffers as a result. But note, many other times, the automation *does* work just fine. And that is why it is there.
QUESTION 2
========
The measure you seek can only be done by visually comparing the result of the scan with the actual negative. You can't put a number on it. For example, look at areas of the sky, or shadows. Do you see excessive noise (grainy when it shouldn't be, or blotchy when it shouldn't be)? This kind of problem can't be seen in a histogram or in file sizes, because it represents small errors in the value of single points in the image. Pellicle's workflow attempts to assure that the quality - and I emphasize quality, meaning accuracy - of the individual pixel values loaded into Photoshop is as high and usable as possible.
You might find that for an image you scanned with the automation in the scanner, it came out just fine - spectacular. Great, you win! You may find at other times that the image just doesn't seem as good as it could be, when comparing it to the negative under a loupe. In that case, you might try Pellicle's approach, to see if maybe you can squeeze out a better result by taking more control over the scan.
Also, note, and this is important: it is highly unlikely that you will be able to tell much of a difference in any of the above at web resolution. To see the issues, you need to look at your images full size, not scaled down to fit the screen (in Photoshop, choose "Actual Pixels" in the view menu). If you never intend to show your images at the full scanned size, but instead at greatly reduced web size (for flickr, facebook, etc.) most of these issues are moot. Scan in the easiest way that works for you, the results should be perfectly fine, except for tweaking colors a bit to make them look good enough.
-------
By the way, in their defense, I think Pellicle, et. al. did respond to these questions, but the language they are using, relating to integer math, curves, etc. may not be something you are familiar with. I hope the above helps, regardless. I also think the discussion about whether the scanner speeds up or slows down to accomplish all it does is a distraction. What matters is what it produces, and how well you can use that result.
| Photrio.com contains affiliate links to products. We may receive a commission for purchases made through these links. To read our full affiliate disclosure statement please click Here. |
PHOTRIO PARTNERS EQUALLY FUNDING OUR COMMUNITY:  |