Stitching & Interpolation - when is enough enough?

I started an experiment. Originally I wanted to get a wall mural 8x10

feet of the Boston skyline, using a 35 mm DSLR (20D with 200 mm 2.8

prime). On a trip to New York took a lot of images to stich together

as a dry run. Basicaly, a set of 7 rows 10 images each. Decided on

taking the middle 4 rows and stitched them together. This is a much

longer pano than the Boston skyline.

 

Straight out of the 40 image stitching process I had a 3 x 9.5 foot

250 mega pixel image. After several iteration of interpolation I ended

up with 8 x 25.5 foot image at 240 dpi/ppi. The quality was actually

much better than I had expected. I could propably go even larger.

Checked with some printers and this is a $1k+ print job and thats

inkjet with dye inks.

 

So, this is my question, what's the largest image that you've created

digitally? Also, at what point is this simply an excerise to do it

just because you can?

Tell us more about the procedure; how much memory, disk space, and rendering time did it take? What software did you use?

More on many pixels... http://www.gigapxl.org/

To get Pano Factory to take more than 8 frames requires the 64 bit version. This required getting a 'Trial' version of Xp Pro x64 and creating a dual boot machine. The out put from the orignal stitch was 28k+ by 9k+. My old version of PS could only hande up to 30K on any side. So, that required getting PS CS2 which I installed on the 64 bit OS. I've got a fairly powerfull PC an AMD Athlon 3800+ dual core with 3 gigs of memory, 95% - 2.6 gig allocated to PS.

 

The last few cycles of interpolation and sharpening were pushing even it's limit. They processed for over twenty minutes each and the temporary scratch file was over 72 gig. There was nothing, and I mean nothing else running. Every single program in the startup was canceled while this was running. These were the only two apps installed on the 64 bit OS.

<cite>Also, at what point is this simply an excerise to do it just because you can?</cite>

 

<p>When you create an image so large that any practical use for it could be satisfied with a smaller image. For instance, if you want a 1m-wide image to hang on the wall, creating a file which (at a reasonable ppi level) is 1m wide is fair. If you instead created one which is 5m wide, you did it just because you could. Or if it's economical to print a 1m-wide image but not economical to print a 5m-wide image, then a 1m-wide image is reasonable, and a 5m-wide image is just because you can.</p>

 

<p>Kinda like using a medium-format camera with a 40 megapixel digital back to take a passport photo.</p>

Your computer will tell you when enough is enough. In your case, I think it already did!

I'm no expert, but wouldn't it be simpler to handle the stitching via the printer itself? In other words, print multiple images and basically have the printer run continuously without any space between one job and the next.

 

The drivers might need to be tweaked, but I don't see any real tough problem in ensuring that registration is accurate.

 

The images could be then be very narrow (but tall) strips if you had limited processing power.

 

<two-bit random train of though mode on...>

 

Do all the work and adjustments you need to a lower-resolution copy of the full image, save them all as a PS action, and then run the action against all the strips prior to output.

 

The only hassle I can think of would be that you need to keep a little overlap or bleed area on each strip, so that when you make an adjustment that affects neighbouring pixels (eg, blur or sharpen), you'll not be doing it right at the edge of the strip. Then crop that part off as the last step.

The operations needed to convert each image to a section of the panorama are complex; light fall-off compensation, perspective correction, and cropping the overlap. How would that work solely in the printer driver?

I have stitched over 20 4meg (12 meg files) images together using Photoshop 7. I was on a tall building and walked a few paces at each shot so that I ended up with a full 360 deg panarama. I had to join each shot manually because of the perspective change and this is what took the time. The result was not perfect but very few people who saw the end result saw the missmatches. You really need a lot of time to do this sort of work. BTW my pc was standard with about 512meg ram, there was a lot of time waiting for the images to merge etc. Quite often I worked with only a few of the images at a time to help reduce the load on the ram/hard drive.

These guys have done some impressive work for Clyde Butcher.

 

http://photolabofnaples.com/

 

They can make a 6' by what ever length archival-canvas print which can be put on stretcher bars (like an oil painting) for less than a conventional paper print and framing combined.

 

If you have the room, why not make the largest print you can. While it may be just for yourself, others will be impressed. I've always liked the phrase, "too much is just enough."

I find that 12x36 inch prints are a saleable size - large enough to make an immediate impression, small enough to hang in an home, and which fit a standard sized frame (11-3/4 x 36). This takes no more than three D2x images (total 200 MB), four Hasselblad image (total 2 GB) or four 4x5 images (total 4 GB), using a 20% overlap. I use Panorama Factory for the stitching, and take the usual precautions of leveling and pivoting on the front node.

 

A single 4x5 inch negative is equivalent to nearly 30 20D images simply in image area. What's the point of stitching an enormous 2-D array together?

Getting larger fomart performance from your existing equipment. If I were doing it every day or making money at it it would not make sense.

If you interpolate UP too much you just create a huge file that doesnt really make a better image. Some of us printers actually downsize customers images when they are full of bull. They clog up the rips. Many times one can downsize 3 to 30 times in size with no apparent drop in appearance. What is cool is that many times cropped samples made for customers actually in some cases appear better with a massize downsize. Provide your printer with an un upsized input too, it often might be better. Here I have several 36 inch and a 42 inch and a 54inch and a 72" printers. The old 30,000 pixel limit was with older Photoshop versions. Often one hit this limit while making long banners and posters even a decade ago.<BR><BR>A common disconnect wtih many folks nmew to making giant prints is ignoring mounting, shipping and laminating costs, transport and shipping costs while being totally obsessed with interpolation, pixels or micromanaging a printers job.<BR><BR>When dealing with the public some of us printers have giant samples done with a mess of different inputs. Here for film I have 3x4 foot sample boards made with 8mm cine thru 5x7 film, cellphone thru 50 megapixel Phase One digital. The general lay public is usually off an order of magnitude in poster or mural requirements.

Here I manually stick together large images many times. I have many stitch programs too, they dont always work with some times of images well. Some versions have limits in size. Manually stitching was done my many of us printers long before any stitch programs existed. Many times it is really quicker too, after a decade plus one has some secret tricks. I started to do this long ago when pro digitals were only VGA and cost several thousand bucks. When Photoshop added layers the manual stitching tasks became radically easier. <BR><BR>A decent printer should have you sign off on a sample proof section of your giant poster, maybe a dinky 2x3 foot section, so there is no missunderstanding as to quality, colors, hues and tones. With a 72" wide printer this might be a 71 inch wide strip say 6 or 12 inches along the feed direction a cropped section at the same size as the giant poster.

With a huge print the image might suck the light out of your room. Some folks want their images to fit an entire wall in their house or store. <BR><BR>What looks great with a dinky print often will look too dark when the entire wall is covered. The same thing happens with folks new to painting. The small paint chip sample tooks great with alot of light, then when an entire room is painted the walls absorb/soak up the light and the room appears too dark. <BR><BR>We made some wall sized murals for a local sporting goods store and had the same problem. Customers often will ignore any experienced words/preaching and press on with a job that one knows might be too light soaking. A lighter brighter image should be used for a full wall application, unless one wants to add more lighting. It is important for a lab to not get burned with a nonpayment when a bullheaded "expert" customer presses on with an ill conceived giant project. Give them enough rope and they will hang themselves every time. :)<BR><BR>Part of the cost of doing larger prints is customer education, providing samples, the ever constant explaining the endless "how big can I enlarge" questions. Many folks that have never enlarged more than a 8x10 inch size have all these notions learned from the internet, and try to educate printers who have printed 20 foot long images a decade ago. Alot of folks are royally, totally confused about enlargements.

<b>Seaming and panels: </b><BR><BR>With walls that are created from a tiled image, ie several strips or panels, seaming becomes important. How the image is diced up can be important, to reduce the seams. With paper there is shrinkage and expansion with humidity. With with different printers there can be color matching errors, plus linearity of X,Y of the prints. With images created on different days, or prints stored in different rooms, seaming errors can grow. Overlapping should be done like roofing shingles, to make the seam not show as much. Thicker papers also make the seams show more. Seams can be hidden more if done in areas of the images that tend to "hide" the seam. Some folks have no overlap; and just butt up the two seams. If a giant image where there are people, it would be bad practice to seam/splice thru folks eyes.

<I>A single 4x5 inch negative is equivalent to nearly 30 20D images simply in image area</i><P>Try a third that, unless you're a member of the "Kodachrome is equivalent to 25megpixels" club. <P><I>After several iteration of interpolation I ended up with 8 x 25.5 foot image at 240 dpi/ppi.</i><P>My experience with really large prints is you need at least 150dpi of raw image data that isn't interpolated before you start getting into billboard tolerance levels. While I know the large format ink-jet printers aren't miraculous in terms of hardware interpolation, the LightJets I've used do just as good a job as incremental Photoshop upscales. Save a whole lotta time I'll tell you that.

20D = 8.2 Megapixel; thirty times that would be 246 megapixels!. The old Phase One scan 4x5" backs are 35 and 50 megapixel over a 7x10cm image. I doubt a 4x5 image has really more than 100 megapixels of usefull info. A 246 megapixel image would be like a usefull area of a 4x5 say 3.5 x 4.5 inches at 4047 pixels per inch; about double the pitch one normally drums scans a 4x5 at. Thus better guess of a usefull megapixel level is about 60 to maybe 100 for a perfect 4x5" in fantasty land.

I don't known how (or if) this will turn out but in this panarama I stitched about 14 images ( part of the 360deg photo I mentioned earlier). I have turned it on it's side to see if I could display it on this forum as it is now 400 X 2112 pixels.

 

I see no problem other than effort (time) to the size of an image handled in PS. There is probably a limit but I haven't found it.<div></div>

Wow it worked1

Jeff; older versions of Photoshop have a 30,000 pixel limit; new ones have 300,000 pixels. With the older limit it would sometimes bite ypu in the rump. Lets say several years back a customer has this banner image at 3,000 x16,000 and then later wants a print up sized 2x. With older PS versions the upsize would go from 3,000 x 16,000 to 6,000 x 30,000. Note it didnt get to 32,000! There would be no warning; and the aspect ratio would be wrong, and the banner/mural too short. I have seen this happen with images many many yards long; and a yard plus wide. checking for hitting the 30,000 barrier was common with making giant murals and banners.

Thanks for that info. Currently I use PS CS2 and not had any trouble as yet although have not tried pushing it.