Tutorial: Illustrated Guide to B&W scan & processing

[John]

Thank you Buze. I need the type of demo you have presented to work through these steps. My Nikon scanner will rotate at the beginning. Should I wait to do it in photoshop, as you have?

Also, I have never used the layers feature in my Elements 4. Do you have any suggestions? such as just keep doing it till you remember how? 😀

 

[Buze]

John nope the rotation can be done in the scanner program if it does it.. Epson Scan does not do it when using the "raw preview" that is why I do it in photoshop later...

 

[varjag]

Buze, unless I missed something, you save the original scan in JPEG. While I see your point of file size, JPEG is invariably lossy compression standard, and making intermediate saves in JPEG will inevitably lose you some detail, no matter how much one twists out the "quality" slider in save dialog.

I'm sure you were aware of that, so maybe for hires MF scans this loss might be unnoticable, but this is an issue people should be aware of.

But, very interesting text overall. Thanks for putting it up.

 

[Buze]

Modern JPEG compressors are massively better than older ones, I defy you to find any artifact in a high quality JPEG file generated from a raw scan.
Furthermore, any changes would be subpixel in size since the image is reduced by half afterward.
/recompressing/ a JPEG is bad; the quality plummets immediately; but in our case the JPEG was written from a nice, clean RAW scan so it compresses beautifuly, and cleanly.

So you would need s x10 loupe and examine closely a 20x20 print to be able to see anything; that is if you knew there was something to look for in the first place 😀

So all in all, considering the difference in file size compared to TIFF, I gladly give up a /potential/ subpixel here and there 😀

Of course a better format would be ideal, if scanner software supported JPEG2000 I would use it !

 

[rxmd]

Hi,

can't you do a colour scan, then do colour separation in Photoshop, convert the channels to B&W separately and blend them in some sort of weighted fashion? That could give you an extra 1.7 (or so) bits of data on average.

Philipp

 

[varjag]

Modern JPEG compressors are massively better than older ones, I defy you to find any artifact in a high quality JPEG file generated from a raw scan.
Furthermore, any changes would be subpixel in size since the image is reduced by half afterward.

Buze, I like to entertain myself with idea that I know a thing or two about JPEG, since I wrote a CCITT/ITU T.81 conforming baseline encoder/decoder in the past. There is only so much room for improvement within the bounds of JPEG standard, and it is not possible to overcome inherent lossy process of DCT/quantization process that is at its heart. The effect of cosine transform is not localised to single pixels, it is a way to record spatial frequencies and discard the least important in convenient manner. Subsampling can only somewhat lessen it but not avoid.

As I said with your requirements above might not be a problem. It also wouldn't be a big deal when scan resolution sufficiently exceeds resolution of finest detail captured in the image. One simple test to evaluate information loss would be to have a layer of original scan over a layer of jpeg handled one, and substract.

 

[MCTuomey]

Thanks for a very fine contribution to the forum. Would you object if I printed a copy for my personal use only?

Mike

 

[jano]

can't you do a colour scan, then do colour separation in Photoshop, convert the channels to B&W separately and blend them in some sort of weighted fashion? That could give you an extra 1.7 (or so) bits of data on average.

Kids, don't this this on an old computer at maximum quality settings. I did... once... poor computer nearly blew up on me! 😱

 

[Buze]

varjag: my point is that even a slight degradation of source signal is counterbalanced that a gain on the signal/noise ratio /and/ file size. Once again, if I could use something /better/ than JPEG, I would; but my current conclusion is that it is STILL better than a 2400dpi TIFF.
My system gives you a good 10 bits of signal with 1/4 the noise ratio of a "16 bits" scan. Knowing that 16 bits scans are NOT 16 bits scans as soon as you move the black/white points in the histogram (with a run of the mill negative, you get about 1/2 or 2/3 of that(, and that Photoshoo will eat one bit of ANYTHING that is not 8 bits too...
Mike: You are welcome !
Philipp: Not sure, /theoricaly/ yes but practically it's doubtful since most scanner noise/signal ratio differ wildly from channel to channel. With some scanner tho, it could get amazing results!

 

[raid]

Thanks for fine write-up and the useful tips.
I am very amateurish with digitasl imaging.
I often use an S-curve, as you have suggested.

How does the S-curve actually work in PS?


Raid