scanning with dslr? part III Extracting gigapixes in your darkroom

[VictoriaPerelet]

Continuing previous thread on using DLSR's to do scans, but changing location of the lens Thought about it awhile ago, but haven't time to try.

Here is a way to extract serious gigapixels in the dark room by using enlarger & DSLR. This setup pretty much passes Imacon & seriously can compete with high end Linotype scanners.

Setup:
Devere 504 enlarger & Rodenstock 180mm lens (any enlarger will work, with others it maybe hard to focus), Canon 10D DSLR with no lens, enlarger focused on sensor, laptop running Canon Remote Capture software.


Enlargement size 59x47cm (22"x18" ), Canon 10D sensor size 22.7x15.1mm, 3072x2048 pixels.
Final digital image size: ~ 95232x51200 pixels 4.8 Gigapixels. To extract same resolution one should scan 4x5 transparency at 24000DPI.


E6 transparency projected 22"x18" :


E6 scratch transparency used for color/resolutions setup using Koren lens chart and color chart:


Part of enlargement (outlined in red) captured on Canon D10 Sensor (enlarger lens at f16) and 1x1 Crop from that capture:

You are looking at 24000DPI DSLR scan, 4+Gigapixel pic.


I'd guess milling machine table can be good to mount DSLR and move it. Several shots can be made to do HDR by either adjusting exposure or enlarger lens aperture. Color balance can be nicely tuned by using enlarger color dials.

With modern DLSR's (aka 5d mk II) one may get 10+ Gigapixel images that can be printed many meters wide

I'll add more details to note at:
http://www.victoriasphoto.com/Notes/Enlarger_Scan/

[Cesar Barreto]

Hi, Victoria.

It's an interesting concept for sure, but I guess how would you'd plan to stitch this whole set of images. At such magnification much of the frames would only show grain patterns, mainly on out-of-focus and flat areas, and I noticed that PS seems to behave erracticly on those situations.
Wich program would you suggest?

[willwilson]

Cesar hit the nail on the head with this one. For example, a 1.2 gigapixel image taken with a Canon G10 and gigapan robot took me 9hrs+ to stitch on a 3.4ghz P4 with 1.5gigs of ram. I can't even open the stitched file in ps because it crashes my machine.

Here's a link: http://share.gigapan.org/viewGigapan.php?id=19978

I am afraid there would be stitching errors galore with this type of method, not to mention how long it would take to stitch something that large (24-48hrs?).

Interesting idea though, and it's always nice to be reminded how much info you can cram onto a 4x5 neg

[Nathan Potter]

Nice work Victoria. Yes you are focusing on only the problem of image capture - so far - but one thing at a time. And your example is in the gigapixel range. Reducing the capture to a few to several hundred Mpixels will relieve the stitch burden to a large degree and still provide a pretty impressive pixel count (probably more in line with the intrinsic resolution on standard 4X5 film). 8X10 format is more difficult of course. The stitching task is currently somewhat demanding from a signal processing point of view but pattern recognition technology is already highly developed for military ops. that can be translated for this application - in time. I think the major problem would be the Dmax (dynamic range) capability - and that I might attack by making a multiple exposure at each site while varying the intensity of the backlight.

I say again that this is a realistic scanning application that could replace existing scanners provided some engineering development gets done. A dedicated machine including the digital capture device, lens, backlight, and film translation stage and stitching software could certainly be built now and sold new for several thousand dollars or less given enough manufacturing volume.

Nate Potter, Austin TX.

[Gene McCluney]

This whole process is long and labor intensive. A dedicated film scanner makes scanning easy by comparison. Other than the giga-pixel resolution (which I can't see using) what advantage is there?

[Nathan Potter]

This whole process is long and labor intensive. A dedicated film scanner makes scanning easy by comparison. Other than the giga-pixel resolution (which I can't see using) what advantage is there?

Gene, almost no advantage currently when attempted manually. But there is a bit of history here which got at least me thinking a bit.

At the most basic, if you are poor, and need to have a few high res scans of LF images you could certainly get some using a digital camera and a lot of labor as I think was the intent or the hope of the original poster.

But then there was a bit of discussion about the lack of new high quality drum and flatbed scanner development directed toward film scanning. There is a thought that since digital capture is now so dominant we may not be able to buy reasonably priced high quality scanners much longer - the market just isn't there to justify manufacture.

So my take, and interest, was to consider what would be involved in digital sensor capture of LF film. As sensors come down in price does it make more sense to capture 10 or 20 Mpixels in a single shot rather than using a single or multiple row
of receptors as now is the case? Two dimensional rather than pseudo one dimensional capture so to speak.

Clearly as you alluded to there is no volume advantage in two dimensional capture unless a fast system is put together including the stitching algorithm. But processor speeds and attendant costs are such now that one could certainly stitch together a Gpixel of data in less than 1 minute (if you ever really need that much).

The larger issue of course is any potential market to support volume manufacturing. If current scanning development is waning then why would one expect a sensor based capture system to succeed any better. A good question.

What this post is doing is just fooling around with ideas.

Nate Potter, Austin TX.

[Cesar Barreto]

On the previous thread I mentioned being doing some trials, but I did actually decided to use this technique for a whole exposition to be open on early may.
There are some trades off and I would really like if Nikon made a large format scanner as good as 9000ED model. But I have two flatbed scanners and neither would save me on this situation, as the Microtek i900 is just usefull for proofs and my old Cézanne means more trouble than I can afford to deal with.

There's no doubt that a good scanner could make my life easier, but I can't think about ever buying a drum scanner and even the best flatbeds are scarse and badly supported here in Brazil. So I think this low-tech and low-budget route stands it own merits and deserves a close look.

If, for instance, those people who projected the Gigapan machine wished so, I think they could easily project some hardware to make tens ou hundreds captures from large format frames and all the necessary processing voodoo to make it work out fine.

For ourselves, maybe things won't be so easy and limiting factors as precision movements and short range sensibility from ccd sensors stop us from dreaming too far. My experience showed differences between frames wich were not supposed to exist and I think that playing around with multiple exposures or HDR would only make stitching a bit more complicate.
But, anyway, every time I see those beautiful panoramas Nasa delivers from Mars, each completed after hundreds captures made on different days and through 4 or 5 cutting filters, I can see that the technology and algorithms already exist and it just a matter of bringing it to our desktop.

Maybe we will see it one day. At least, I hope so.

[Ben Syverson]

4.8 gigapixels from 4x5 is egregious overkill in the oversampling department.

As you can see from the crop, individual grain clumps are rendered at least 20 pixels wide! You can even see the air bubbles in the emulsion (the once-mysterious "pepper grain"), rendered with stunning clarity. The question that leaps--no, bounds--to mind, is "why?"

To capture an entire 4x5 frame and all of its actual resolution, you should only need 12 exposures (4x3) on a 21 megapixel body. So it doesn't have to be so labor-intensive... Sure, it's not as convenient as a scanner, but the quality will blow any scanner away.

[VictoriaPerelet]

This whole process is long and labor intensive.

"long and labor intensive" is synonym for darkroom. Darkroom is where typically unique things get created.

When Kinkos will do multi gigapixel scans value of end product will be at mass market consumer level.

Detail reproduction at resolution posted above is at pair with ink jet level of details printing on transparency (see my other notes). Hand made multi square meter contact print on hand coated surface surely can be unique and yes very labor intensive. Use your favorite alternative method or even liquid light.

Our work is editorial and usually is not reproduced on anything larger than 10x12 - editorial magazine spreads or covers is where most of our work ends up. But we do get collector requests now and then.

Stitching algorithms – google is your friend. "Dr. H. Dersch" is maybe good keyword to start with. Again mostly opensource research development work that is far away from click and drag ease of Photoshop/Kinkos.

[jb7]

I'd guess milling machine table can be good to mount DSLR and move it.

A kind of an etch-a-sketch kind of a thing?
You could even imagine that motorized, with matching software-


In the close up detail, are we looking at debris on the sensor,
or are those dark dots in the image?

Interesting project-