Untitled Document

I find myself in the unfortunate situation of having to scan and work with a previously printed photograph for a new book cover. The original 6X9 transparency has been lost. All I have is a grayscale scan at 150 Dpi, plus a new scan of the printed book cover, which of course will reveal the dot structure created by the printing press.

It is impossible to do a reshoot of the photograph.

What can I do with plate blending, blurring, sharpening, etc. to minimize the damage, so that I can use this image for a new book cover? How much can be achieved through these methods to recapture what I have lost?

Also, if there is anyone in the group who is an expert at this, I would be happy to pay to have this done, if you reply to me offline at sambiser@mac.com

Sam Biser

___________________________________________________________________________

If you can get access to the original film separations they can be rescanned on a Creo Eversmart Supreme and recombined as a DCS file without any noticeable loss. This file can now be used as you lease and even edited. Chances are this was probably printed direct to plate and you?ll have to rescan the book cover as you stated and using a defocus in the capture. I?d try several different settings to get the best reproduction without too much softness.

Gary Roushkolb

___________________________________________________________________________

Sounds like a job for LAB.

Open the rescreen scan and convert to LAB. Place the grayscale scan in the LAB channel (obviously you'll have to distort it carefully to fit) and use this to blend away the dot from the L channel. Use various blurring, dust and scratches, median, noise ninja, etc to get rid of the pattern in the A and B channels without getting color leaks.

How big of a file do you need? How big of a scan is that grayscale?

--

J Walton

___________________________________________________________________________

The grayscale scan is 1.7 megs. The image needs to bleed off the front cover of a 6x9 book, and needs to be as high-quality as achievable, so I don't know how big the final color file should be. -- Sam Biser

___________________________________________________________________________

Hi Sam:

I have used a PS plug-in product called Intellihance Pro 4, originally sold by Extensis, and not licensed to OnOne Software.

http://www.ononesoftware.com/detail.php?prodLine_id=3

The method I have used is to scan the printed page at 600 dpi. Edit the file in PS at 200 to 300% to clean the scan and remove any artifacts. Once you have a clean file, apply the Intellehance Pro plug-in. Select Descreen - the drop down lists various printing methods. I have found that Newspaper works best. Return the file to PS and apply additional tweaks - Levels - curves - finally sharpening USM, just enough to define the edges.

Using this method I have produced images that look better than the printed page. You can download a 30 day trial to try it out.

Regards,

Mike

Fine Photography

by Michael Demyan

www.mikedemyan.com

610-758-9769

___________________________________________________________________________ .

Thanks Mike. Do you think it will make some difference if the printed book cover is scanned in at 450 dpi, as they were planning, or at 600 dpi? Sam (540) 456-8184

___________________________________________________________________________ .

If you post sample images, I'll take a crack at it.

The 150 dpi scan probably has more detail than the book cover, so my first tack would be to take the luminance info from your scan, and color info from the book cover. Use your 150 dpi scan of the transparency as the L channel of Lab (with curving). Scan the book cover, convert it to Lab, blur the a and b channels and use them as the a and b channels of your final image.

Otherwise, I like FFT / IFFT for removal of halftone dots. The results can be dramatic. Here's an article describing how to do this using a free windows plugin - sorry no Mac version.

http://www.curvemeister.com/forum/index.php?topic=62.0

For color, convert to CMYK (experiment with the GCR setting that gives the clearest dot pattern in each channel) and process the channels separately with FFT, then recombine for a dot-free version.

Mike Russell - www.curvemeister.com

___________________________________________________________________________

Mike, thanks for the link. FFT/IFFT seems to work quite well, but I am not sure I understand how best to make it work and what portions of Red channel to touch up after running the FFT plugin. Any thoughts on how to get the most from this thing?

Lou Dina

___________________________________________________________________________ .

Lou, I used FFT to remove a baseball net in a thread on DigitalDrin. The first image only had half the net removed, and the second smaller image removed both vertical and horizontal frequencies where the net resided:

http://www.dgrin.com/showthread.php?t=64775

Some very good information on FFT can be found here:

http://retouchpro.com/tutorials/index.php?m=show&id= 185

http://www.retouchpro.com/forums/ima...olf-balls.html

http://www.pages.drexel.edu/%7Eavc25/archive.htm#FFT

http://www.skeller.ch/ps/fft_action.php

http://rsb.info.nih.gov/ij/docs/examples/FFT/

As well as FFT, "denoising" and or "inpainting" are also good options:

http: //www.greyc.ensicaen.fr/~dtschump/greycstoration/img/res_moule.html

Regards

Stephen Marsh

___________________________________________________________________________

Thanks, Stephen.

That helps quiet a bit. It really is amazing. Great tool in the bag of tricks.

Regards,

Lou Dina

___________________________________________________________________________

Hi Lou,

Stephen has provided some good links. I would add a recommendation to experiment, concentrating on bright off center "star" spots, and making changes incrementally until the dots are gone.

Mike

___________________________________________________________________________ .

Hi Sam,

there are many techniques that allow good results for this kind of problems, in some cases better than the original.

I don't know Intellihance but in my experience you can have good results already in Photoshop. A good starting point, as Michael Demyan said, is to scan the printed page at a resolution of 600 ppi or at least at the double of the final output resolution, of course at 100% scale.

It's important that the scan is very good, with the book perfectly straight on the scanner to avoid the little rotation of the printed page. Now you'll have a very big file that allows to you many moves.

As you talk about a grayscale scan I suppose that the image is B&W. This make less easier to eliminate the moire but not so much. Anyway let's suppose it's a B&W photo (for the color I can give you additional suggestions)

First of all you have to evaluate some things before deciding what to do. Which kind of subject is it? Is it a landscape, a beauty or a still-life? How much detail does it have and how much do you need to keep? And in which part of the image the moire is more visible?

These factors are critical for the amount of blur you'll have to do.

In most of the cases you can try to do a Gaussian Blur raising the radius until you'll see no more moire but you will keep all the details you need.

In certain images you can also find useful the Surface Blur that holds the edges or sometimes the Smart Blur.

If you're using Photoshop CS3 I would suggest to convert the layer into a Smart Object. After this you can apply all blur filters combination to see what happens to the image and you can always change the settings and their order as well to decide the best conbination of them for the blending mode as well. You can also reduce the opacity of any applied filter.

In some cases the Dust & Scratches filter also could help, you'll have to try it.

After you find that the image has no moire at all but has a sufficient detail, it's time to resample the image to the final resolution (300 ppi or what else) setting the Interpolation method in the Image Size dialog to Bicubic Sharper to have a slight sharpen before taking care of the final sharpen for restoring the detail.

It's very important that you really don't see the moire before sharpening because once the moire is gone you can sharpen as much as you want, the moire will never come back. It's better to have an image with a little blur instead of the moire revival.

I hope this will help you.

Alessandro Bernardi

___________________________________________________________________________

Thanks, Mike.

I have played with FFT/IFFT some and it is really quite interesting and effective. Remarkable what you can do with math! Thanks again for the great link to this plugin. I already used it to clean up a scan of an 1890's era photo on heavily textured paper. Worked great and made the rest of the retouch a lot easier.

Lou Dina

___________________________________________________________________________

Are there any options for handling FFT in OS X, preferably in Photoshop?

I'm just checking out an option for GIMP, and will report back with results (after the weekend).

Rick Gordon

--

___________________________________________________

RICK GORDON

EMERALD VALLEY GRAPHICS AND CONSULTING

___________________________________________________

___________________________________________________________________________ .

Alessandro Bernardi wrote:

Another method makes active use of rotating the original on the flatbed, to reduce moire in between the dots and the scanner array.

My understanding is that there is a 150 ppi grayscale *contone* scan of the original colour film, plus, Sam can scan the printed book as RGB.

Sam has not mentioned if he wishes to remove text and other elements from the book scan, so that it is like the grayscale scan of the image - which I presume is at the same size, but with bleed and no text etc.

Can you comment on the above Sam?

Stephen Marsh

___________________________________________________________________________

That is something I never thought of -- making sure the book cover image lays straight on the scanner.

If I use the rotation method, how much rotation (approximately) does it take to reduce the moire in between the dots and the scanner array?

Regarding Stephen Marsh's question: I do not need to eliminate text from the image, as the text is on top and is outside the image area -- and it is quite easy to remove it to blend -- if needed -- with the quality grayscale scan I have of the color image.

A question of my own: How important is it to get the scan at twice the final output resolution of 300 dpi? The service bureau just handed me a scan done at 400 dpi, and said that was ample to reveal the dot structure of the printed image -- and that appears to be correct. Should I have them redo the scan at 600 dpi?

Thanks.

Sam Biser

(540) 456-8184

___________________________________________________________________________

Rick Gordon wrote:

There is ImageJ for OSX, but it falls over when FFT filtering moderate size files.

The folk at www.reindeergraphics.com offer commercial quality FFT filtering (not lossy/noisy like many free FFT filters). Their server is down for me at this time, so I can't be more specific. It is in FoveaPro, but that is not meant for graphic arts or photography and is expensive. I don't know if it part of Optipix, but if it is then that is a less expensive option.

There may be others, I have not found them though, so I too would like to hear about OS X FFT, either stand alone or for Photoshop.

Of course, Intel Mac users could fire up MS Win OS if they have the various software required!

Another MS Win FFT option here (a regular GUI approach to FFT):

http://www.descreen.net/eng/soft/descreen/descreen.htm

Stephen Marsh.

___________________________________________________________________________

Sam I apologize for my English understanding (and writing), but I think it's time to clear some points.

1. Does the scan from the service bureau has the moire visible or not?

2. The scan is in color or grayscale? And in which color space, RGB or CMYK?

3. What does it mean "was ample to reveal the dot structure.... and that appears to be correct"? It means that you can see the printed dots but they don't cause any moire? Correct is referred to what?

4. Do you have to print the scan at 100% scale of the original or you have to enlarge it?

All these points are critical for deciding what to do and in which way.

When you get a scan from a printed image you have generally the first problem that is the interference between the halftone screen and the pixels of the image, the moire.

With some scanner softwares you can eliminate the moire with a function called descreen but you may still see the halftone screen (the printed dots). It depends also on the frequency of the printed halftone screen. In a page printed at 133 lines per inch you will see much more the dots than in a page printed at 175 lines per inch.

So I think the target is to have no moire at all and to eliminate the halftone screen as well.

In both cases you'll need a scan at a resolution that allows to blur a lot the dots without destroying the details. At 600 ppi and 100% scale you can really apply strong amount of different kinds of blur with the technique I've mentioned before and when you resample with a strong percentage the image, the resample algorithm recompact the pixels eliminating the defects.

You can try both the Bicubic Smoother or Bicubic Sharper before the sharpening and see how good is the final result.

If you say that the service bureau has made the scan at 400 ppi, in my experience when going from 400 to 300 ppi the difference is about 30%, it's not high enough and the strong blur will affect the detail.

Another problem is that if you print the image at more than 100% scale you will certainly see the dots so, again, you'll need to blur.

But, in any case, once the file becomes a normal image with no moire and no printed dots, it could be sharpened as much as you like, and the moire and the dots of the halftone screen will never appear again.

Of course deciding how much blur is needed depends on how large is the halftone screen, so you'll have to evaluate the result zooming in and out in Photoshop and seeing when you'll notice the moire.

In most of the cases if there is moire you'll see it when viewing the image at 33.3% or 66.67% or 16.7%. The intereference betweeen the pixels of the image and the pixels of the monitor become very critical at these percentages so the moire will immediately appear.

This doesn't happen at 12.5%, 25% or 50% because Photoshop divides the pixels of the image by 2 or a multiple of 2 to show the image on the monitor and 100 divided by 2 or 4 or 8 it's always better than divided per 3 or a multiple of 3 or any other multiple different from 2.

And this is also why I suggest to downsample from 600 to 300 ppi, the result will be that two pixels will be mixed into one. Otherwise 400 divided per 300 ppi will give a periodic number so some errors will be introduced in this calculation and the pixels will not be divided equally. Sometimes is better and sometimes is worse.

The trick suggested by Stephen goes in this direction, use the rotation managed by Photoshop to mix the pixels together therefore sometimes works well sometimes not so much, it depends also if the page is printed well or not and if the registration of the printing inks were good or you can see some out of register in the print.

However the principle is similar as when you rotate the image in Photoshop the single pixels of the image will be modified and in a certain sense blurred.

I think also that is suitable if the scan is made by you and not when is done by a service bureau.

In addition to all these considerations, if the image is in color, then there are many other ways to eliminate the moire and the halftone screen that involves some moves on the single channels of the image in RGB, some in LAB and so on.

So, let's make a step at once and see what you need.

Alessandro Bernardi

___________________________________________________________________________

.

Sam Biser wrote:

It is often a good thing to make sure of, but in the case of repeating patterns, sometimes it is not the best thing to do. It may be dependent on the scanner, resolution settings, original image content or other factors. Sometimes the straight scan will capture better halftone dots, if this is desired. Other times the angled scan will provide better results and not resolve the paper between the dots as well, but still provide some detail (dots).

If one scans straight and there is a moiré in the composite, then angling the scan on the glass and then rotating it back in Photoshop via the crop tool or another method may offer better results. often, even if there is no moiré visible in the composite image, the angled scan will offer better individual channel content, often in the blue.

Sometimes the scanner area and the size of the origianl will limit the amount of rotation available. Dan generally recommends 45 degrees for CMYK halftone originals (but it may depend on the dominant hues) and 15 degrees for monotone originals using a 45 degree halftone angle.

I have uploaded a crop of an old image from an Umax flatbed at 600ppi optical res. with the left half being a regular 90 degree angle scan and the second being a 45 degree angle scan that was rotated back to true in Photoshop (no other processing done, no blurs, median etc):

http: //tech.groups.yahoo.com/group/colortheory/files/descreen-examples.jpg

Best,

Stephen Marsh.

___________________________________________________________________________

.

"Alessandro Bernardi" wrote:

Moire is not visible

Colorspace is Adobe RGB

No moire.

Printing at 100% scale

Sam Biser

___________________________________________________________________________

.

J Walton wrote:

Or for RGB...

One can split the file into separate luminosity and chroma components, as recently introduced in this thread:

"accepted wisdom" - Blend If, "Gray"

http: //tech.groups.yahoo.com/group/colortheory/message/19031

Similar initial thoughts to my own...except for using Luminosity and RGB instead of Lab mode.

I guess by the lack of response from the majority of our 3,500 members, that nobody took things any further with the copious information provided by the www.freegamma.com website? Do you agree or not with the Lobster marketing points (which any user can take advantage of without using Lobster)? Working with RGB luminosity is currently one thing that I am very interested in, even if the ACT list is not!

Sincerely,

Stephen Marsh

___________________________________________________________________________

Stephen,

I am interested and was hoping to see further discussion on this topic. After reading your initial post, I was a little in the dark as to exactly what you were talking about and was hoping to see more discussion on the topic in hopes of getting a better grasp of the points you were trying to make. So I for one would like to know more. Anything to advance my knowledge on the topic.

Sincerely,

John Arnold

___________________________________________________________________________

My first impression is I'd rather work with it as a color space (channels).

Lori Sabo

___________________________________________________________________________

Lori Sabo wrote:

This is indeed the conventional approach that may first spring to mind Lori, although it does not advance the discussion on working with RGB Luminosity and chromaticity data, which I previously promised to do if list members would meet me half way and experiment and make active contributions to the discussion thread. The moment seemed appropriate with Lab mode lightness being proposed as one stage of the work-flow of the restoration of the full colour halftone scan with the smaller/lower resolution "original derivative" grayscale con-tone data.

Regards,

Stephen Marsh

___________________________________________________________________________

What is the unique power in working with RGB luminosity and chromaticity data, as opposed to the

conventional approach of working with the LAB lightness channel to restore a full colour halftone scan.

What can this approach do better than LAB?

Thank you,

Sam Biser

___________________________________________________________________________

Hi Sam, well - I think your colour descreen image makes a good case in point! <g>

Here is the original post that kicked off this stalled topic:

http: //tech.groups.yahoo.com/group/colortheory/message/19027

I wrote:

"Perhaps some list members would like a civil discussion where we all learn some new things about basic Photoshop operations, that may go against the "accepted wisdom"? If so, it will require that people meet each other half way, the topic will not progress unless there is feedback that advances the discussion (otherwise I would simply make a post, rather than make such an invitation which requires work on the part of participating members)."

I don't enjoy one-sided conversations, which is why I invited the ACT list to explore and discuss the points.

Or put another way, what one will get out of the discussion depends on what one puts into the discussion (come on, meet me half way guys!).

Regards,

Stephen Marsh

___________________________________________________________________________

Stephen.....

I have a minuscule percentage of experience than most of the other members of this group but I have an avid interest in all things regarding digital imaging and color theory. I have read and studied the information on the freegamma.com web site as well as from a couple of other similarly inclined sources. I do understand the discussion on the benefits of the luminosity layer but am much more in the dark on the full utility of the separate chromaticity layers.

I would welcome the chance to participate in a discussion on this topic that purports to go against "accepted wisdom".

Russell Cowgill