Scanner (in)accuracies

Graphics amateur here, but I stumbled across something which is probably old hat to you pros. However, I’d like to share it with the class and see what you lot think about it…

I’m currently agonizing over scanning resolution and bit-depth to use to scan irreplaceable family photos, negatives and slides. The idea was to use optical maximums for archiving and then, at a later date, possibly scale down for printing and displaying, as required.
However, 48-bit color at 2400 dpi produces obscenely huge files so I devised a little test to see how accurate those specs really are.
I scanned an image twice in succession. A binary comparison of the two files revealed them to be quite different. And it doesn’t matter what resolution or color depth I used. Even 1-bit, 50 dpi scans come out different!?

Puzzled, I exported the images as RAW and wrote a quick-and-dirty program to average them out. I used 5 scans, added up values for each pixel and then divided by 5 producing a simple "average" image. This is a very pedestrian, plain-vanilla algorithm for data acquisition, I didn’t even bother eliminating the extremes.

(BTW, how do I average images in PS 6.0.1? I played with Apply image and Blending modes but I just can’t produce the desired effect. It’s all those options. Brain hurts…)

Anyway, back to my "average" image. Instead of improving things, this actually made matters worse as the resulting image came out "soft" and out of focus.

So, I wrote another little quick-and-dirty "difference" routine to compare two images like so: if file 1 pixel < file 2 pixel I assigned it a value of 255 ("white"), if file 1 pixel > 2 it became a 0 ("black"), and if both are same the resulting pixel became 127 ("gray"). Once I loaded this "difference image" it looked like it was put through an emboss filter and it had R, G & B "shadows" indicating both vertical and horizontal displacement.

(Again, how would I produce such a "difference image" in PS 6.0.1?)
Anyway, this means not only are room-temperature CCDs quite inaccurate but stepper motor tolerances appear quite bad too.

Sure makes me wonder about all those "generous" bit-depth and resolution claims, and puts my agonizing over them into perspective.
Don.

I would doubt that you’d be able to see, on screen, any difference between a 300dpi scan and a 2400 dpi scan except for size.

I think you’d need to decide how you want to output your images when you Unless you’re planning poster size images I think
300 or even 600 dpi would be plenty.

Graphics amateur here, but I stumbled across something which is probably old hat to you pros. However, I’d like to share it with the class and see what you lot think about it…

I’m currently agonizing over scanning resolution and bit-depth to use to scan irreplaceable family photos, negatives and slides. The idea was to use optical maximums for archiving and then, at a later date, possibly scale down for printing and displaying, as required.
However, 48-bit color at 2400 dpi produces obscenely huge files so I devised a little test to see how accurate those specs really are.
I scanned an image twice in succession. A binary comparison of the two files revealed them to be quite different. And it doesn’t matter what resolution or color depth I used. Even 1-bit, 50 dpi scans come out different!?

Puzzled, I exported the images as RAW and wrote a quick-and-dirty program to average them out. I used 5 scans, added up values for each pixel and then divided by 5 producing a simple "average" image. This is a very pedestrian, plain-vanilla algorithm for data acquisition, I didn’t even bother eliminating the extremes.

(BTW, how do I average images in PS 6.0.1? I played with Apply image and Blending modes but I just can’t produce the desired effect. It’s all those options. Brain hurts…)

Anyway, back to my "average" image. Instead of improving things, this actually made matters worse as the resulting image came out "soft" and out of focus.

So, I wrote another little quick-and-dirty "difference" routine to compare two images like so: if file 1 pixel < file 2 pixel I assigned it a value of 255 ("white"), if file 1 pixel > 2 it became a 0 ("black"), and if both are same the resulting pixel became 127 ("gray"). Once I loaded this "difference image" it looked like it was put through an emboss filter and it had R, G & B "shadows" indicating both vertical and horizontal displacement.

(Again, how would I produce such a "difference image" in PS 6.0.1?)

Anyway, this means not only are room-temperature CCDs quite inaccurate but stepper motor tolerances appear quite bad too.

Sure makes me wonder about all those "generous" bit-depth and resolution claims, and puts my agonizing over them into perspective.

Don wrote:

…
(BTW, how do I average images in PS 6.0.1? I played with Apply image and Blending modes but I just can’t produce the desired effect. It’s all those options. Brain hurts…)

Scarecrow puts finger to head, and says quickly: "Calculate the average of

set of images by copying each image to a layer whose transparency is set

1/2 that of the previous layer."

(BTW, how do I average images in PS 6.0.1? I played with Apply image and Blending modes but I just can’t produce the desired effect. It’s all those options. Brain hurts…)

Scarecrow puts finger to head, and says quickly: "Calculate the average of a set of images by copying each image to a layer whose transparency is set to 1/2 that of the previous layer."

(Again, how would I produce such a "difference image" in PS 6.0.1?)

Superimpost the layers, set difference mode, select the move tool, and use the arrow keys. Or if you prefer do it with channel ops and the displace filter. BTW – you will probably find that no matter where you move the second image, it will not quite overlap the previously scanned one.

Anyway, this means not only are room-temperature CCDs quite inaccurate but stepper motor tolerances appear quite bad too.

You got it – particularly the second one.

Sure makes me wonder about all those "generous" bit-depth and resolution claims, and puts my agonizing over them into perspective.

Again, you got it. Here’s another experiment. Scan a uniformly colored, texture free material, such as a piece of translucent plastic. This will give you a good idea of the noise level of your scanner, how many bits are really being scanned. I think you will find that the marketeers have been at work, and many of the 16 bits per channel are noise.

Kudos to you for actually experimenting, and coming up with new information!

Graphics amateur here, but I stumbled across something which is probably old hat to you pros. However, I’d like to share it with the class and see what you lot think about it…

I’m currently agonizing over scanning resolution and bit-depth to use to scan irreplaceable family photos, negatives and slides. The idea was to use optical maximums for archiving and then, at a later date, possibly scale down for printing and displaying, as required.
However, 48-bit color at 2400 dpi produces obscenely huge files so I devised a little test to see how accurate those specs really are.
I scanned an image twice in succession. A binary comparison of the two files revealed them to be quite different. And it doesn’t matter what resolution or color depth I used. Even 1-bit, 50 dpi scans come out different!?

Puzzled, I exported the images as RAW and wrote a quick-and-dirty program to average them out. I used 5 scans, added up values for each pixel and then divided by 5 producing a simple "average" image. This is a very pedestrian, plain-vanilla algorithm for data acquisition, I didn’t even bother eliminating the extremes.

(BTW, how do I average images in PS 6.0.1? I played with Apply image and Blending modes but I just can’t produce the desired effect. It’s all those options. Brain hurts…)

Anyway, back to my "average" image. Instead of improving things, this actually made matters worse as the resulting image came out "soft" and out of focus.

So, I wrote another little quick-and-dirty "difference" routine to compare two images like so: if file 1 pixel < file 2 pixel I assigned it a value of 255 ("white"), if file 1 pixel > 2 it became a 0 ("black"), and if both are same the resulting pixel became 127 ("gray"). Once I loaded this "difference image" it looked like it was put through an emboss filter and it had R, G & B "shadows" indicating both vertical and horizontal displacement.

(Again, how would I produce such a "difference image" in PS 6.0.1?)
Anyway, this means not only are room-temperature CCDs quite inaccurate but stepper motor tolerances appear quite bad too.

Sure makes me wonder about all those "generous" bit-depth and resolution claims, and puts my agonizing over them into perspective.
Don.

Although CDRs are indeed cheap, according to my rough calculations, even with the above approach I’d still be looking at a minimum of 200 CDs (don’t trust the format jungle of the DVDs yet). And that means my digitized images will be bigger in both volume and weight than the originals – which I found a bit amusing… To be fair, this will shrink down to a handful of "album" CDs of JPGs, so that’s not too bad.

One other thing I noticed in my various tests is that negatives, even though properly stored, seem to suffer much more physical damage than prints. This was not apparent to the naked eye, but once scanned all sorts of things were revealed. On the other hand, although physically in better shape, color prints tend to fade much faster than film.
Thanks again!

Don.

—

On Sat, 26 Jul 2003 22:44:13 GMT, "Jerry Beasley" wrote:

Don,
I’ve been through this exercise before as I did a similar project

boxes of my family photos. There are many valid arguments on appropriate scan size, but as you alluded to, you don’t necessarily know how your

will be used in the future. My blanket suggestion here is go as high resolution as you have time and patience for. CD-Rs are cheap, and as

pointed out, you can reduce the size later.

Although CDRs are indeed cheap, according to my rough calculations, even with the above approach I’d still be looking at a minimum of 200 CDs (don’t trust the format jungle of the DVDs yet). And that means my digitized images will be bigger in both volume and weight than the originals – which I found a bit amusing… To be fair, this will shrink down to a handful of "album" CDs of JPGs, so that’s not too bad.

Just to butt in for a second, DVDs are stable if you use the proper recorder for the proper DVD disks. Prices are getting dirt cheap for the new drive that will record both DVD-R and DVD+R. I thought as you did but I am finding out the DVD, especially the DVD writable disks are so much more stable than CD writable. Prices for DVD disks can be picked up, on sale, for very economical prices, especially the recordable. Either recordable or writable will play on DVD players without a problem. There are even editing software that will allow you to burn photos and play them on your DVD player on you TV. Helps with those special people who do not have computers or DVD player on their computers. Convert the cds aprox 700 meg to the nearly 4.7 gig that you get on DVD and you will quickly see the difference. Recordable DVDs are the way to go if you can fill them up, otherwise you will DVD rewritable’s to add to at different sessions. Please do not dismiss the DVD solution until you have thourghly researched it my discussing with people who have used DVDs. Also, if anyone can offer any additional advise please do so.

Although CDRs are indeed cheap, according to my rough calculations, even with the above approach I’d still be looking at a minimum of 200 CDs (don’t trust the format jungle of the DVDs yet). And that means my digitized images will be bigger in both volume and weight than the originals – which I found a bit amusing… To be fair, this will shrink down to a handful of "album" CDs of JPGs, so that’s not too bad.

One other thing I noticed in my various tests is that negatives, even though properly stored, seem to suffer much more physical damage than prints. This was not apparent to the naked eye, but once scanned all sorts of things were revealed. On the other hand, although physically in better shape, color prints tend to fade much faster than film.
Thanks again!

Don.

—

On Sat, 26 Jul 2003 22:44:13 GMT, "Jerry Beasley" wrote:

Don,
I’ve been through this exercise before as I did a similar project

with

boxes of my family photos. There are many valid arguments on appropriate scan size, but as you alluded to, you don’t necessarily know how your

images

will be used in the future. My blanket suggestion here is go as high resolution as you have time and patience for. CD-Rs are cheap, and as

you

pointed out, you can reduce the size later.

Although CDRs are indeed cheap, according to my rough calculations, even with the above approach I’d still be looking at a minimum of 200 CDs (don’t trust the format jungle of the DVDs yet). And that means my digitized images will be bigger in both volume and weight than the originals – which I found a bit amusing… To be fair, this will shrink down to a handful of "album" CDs of JPGs, so that’s not too bad.

Just a quick note. Format Jungle, Him Dead. He killed by Recorder using both -RW and +RW standards. It no longer problem. DVD now King of De Jungle 🙂

I scanned an image twice in succession. A binary comparison of the two files revealed them to be quite different. And it doesn’t matter what resolution or color depth I used. Even 1-bit, 50 dpi scans come out different!?