penguin logo\
 
  SHOOTING TROUBLESOME ART, part IV: seriously enormous paintings
Scanning a huge painting (10' x 15') in two horizontal rows of 5 scans each in a tight space.

Note: This is a posting for people in the art reproduction business as well as patrons so portions may be too technical for your needs. If you are an artist that paints huge canvases, and have any questions, please feel free to write jean@RonFinleyStudio.com

Where do murals begin and paintings stop?
You don’t need a classics studies degree to know that “mura” is Latin for “wall”, so “murals” are wall paintings, right? At UCLA’s “SPARC” (Social and Public Art Rescue Center) the definition is considerably broader. In addition to the preservation of the urban murals which abound in Los Angeles, “SPARC is particularly committed to producing and promoting work that reflects the lives and concerns of America’s ethnically and economically diverse populations including: women, the working poor, youth, the elderly and LA’s newly-arrived and numerous immigrant communities”. This translates into sponsoring new works, often as painted canvases, usually in the grand muralist styles of story telling, and most always BIG.

One such painting, “La Danza de la Tierra” was done by Professor Judith Baca, co-founder of SPARC, commissioned for a museum in Dallas, Texas. We were recently hired to create a high-resolution scan of “La Danza” that could be printed 100% original size (10’ x 15’) on a to be hung in a new Los Angeles restaurant as a centerpiece.

 
 

"La Danza de la Tierra" by Judith Baca, 2009 acrylic on canvas,
120" x 180"

(To see more of Prof. Judith Baca's work and the work of the Cesar Chavez Digital Mural Lab, please visit them at www.sparcmurals.org/ucla)

"La Danza" by Jutdith Baca, 2009
 
 

SPARC’s hope was that we could scan the work in Judith Baca’s studio, a splendid Frank Gehry-designed box on a canal in Venice. We would have preferred more space, like a nearby movie studio, but this would have involved disassembling and re-assembling the painting one more time than the artist wanted to chance. But how could we create a high-resolution scan of a painting of this size in that somewhat limited (by our standards) space?

We routinely copy large works of art by shooting them in sections (called “tiles”) and assembling them in Photoshop (or more specific applications like Panotools) to generate a “hi-rez” file. The easiest way to do this is to move the artwork precisely along the x axis (horizontally), scanning in pre-determined intervals as we go. In the case of a 15’ wide painting, we would need almost 30’ of left-right travel space to reposition the painting. Ms. Baca’s loft was exactly 30’ deep so this was possible. But how would we control the movement of the painting with the necessary precison?

In our first meeting with Sparc's Carlos Rogel and Farhad Akhmetov, we discussed the use of a track dolly along the lines of our “moving wall” which we use in galleries, but decided it would eat up too much space on the z axis (depth) and force us into using a less-desirable, wider-angle lens. I suggested to Ms. Baca installing an industrial-gauge sliding door track the 30’ length of the studio, so long as she didn’t mind some lag bolts in her nice wall. She liked the idea, seeing the track installation’s potential for other large works in the future. Next, I had to plan out the scan layout to create the “hi-rez” printing file.

What is a “Hi Rez” Scan of a 15’ Painting?
The knee-jerk answer is usually “a 300 dpi file, of course,” an answer that drives those of us who do this for a living nuts. First off, “resolution” is the term used to denote the size of a bitmap file, usually by its pixel dimensions or by the number of megabytes it creates. “DPI” or more accurately “PPI” (pixels per inch) is an output (i.e. printing) specification. For example, a file with a resolution of 2000 x 3000 pixels (6 megapixels) printed at 300 ppi would output a picture 6.6” x 10”. When printed at 100 ppi, it would output a picture 20” x 30”. Both could be considered acceptable “high resolution” under different circumstances.

In the case of copying Ms. Baca’s painting, “hi-rez” meant creating a bitmap document which, when printed full-size, will appear sharp at a near distance, say a foot or so. What will satisfy this? If we were to generate a file to make a 10’ x 15’ print at 300 ppi, it would have to be 54,000 pixels x 36,000 with a flattened file size of 5.4 GB. A file of this big would push the limits of what computers can handle but that is a moot point, as the printing house will accept no file larger than 2 GB. After some quick calculations using Photoshop, we discovered that if we dropped the output 180 ppi, we could create a picture 1.96 GB, making everyone happy. But would it still be high rez? Yes. There might be some small visible difference between 180 and 300ppi, but you’d need to examine them next to each other with a magnifying lupe to see it. Actually, 180 ppi is a very good resolution as it involves no printer interpolation, as this printer (and most others) are 14400 ppi. (14400 ÷ 180 = 8 exactly)

Laying out the scans
Now that we know the target printing output, we can compute the dimensions of the final painting copy file: 120” by 180” @ 180ppi = 32,400 x 21,600 pixels. But how many sections (tiles) will we need to shoot? Our Better Light digital back yields a maximum size of 12,000 x 9,000 pixels per scan. A little quick math tells us that two rows of tiles12,000 high will give us enough coverage for the height of the painting (21.600) including 10% extra image for the overlapping necessary to “blend” the tiles.

But what about the width? Can it be done in four or five tiles? Four looks about right until we superimpose our scale Better Light back image over it. There is very little left-right overlap and would create big problems in assembling the final image. Five tiles, on the other hand, yields generous overlap, an advantage in assembling the image, whichever method we use (footnote).

Two Rows of Scans?
The process of “piecing” together a large file from a lot of small ones inevitably leads to multiple rows of tiles. There are a number of ways to accomplish this feat without having to move the camera (to be avoided) which will be discussed in detail elsewhere. This painting, however, was really simple. We would shoot the lower row of tiles, then flip the painting upside down (as shown), and shoot the upper row of tiles (inverted) from exactly the same position. This solution was made possible by the generous ceiling height in Ms. Baca’s studio and a handy laser measuring device provided by Mr. Rogel.

 
 
Final Tile Layout Plan for Judith Baca's "La Danza de la Tierra" involved 2 rows of 5 tiled scans (12,000 x 9000 pixels each) which would produce enough overlap for assembly, and generate a high-resolution, full-sized print of 10" x 15' at 180ppi.
scan layout for La Danza by Judith Baca
 
 

Lens Selection
We have a large selection of Rodenstock APO Ronar Lenses of focal lengths from 150mm to 480mm. By always choosing the longest one possible, we minimize any possibility of specular reflections. In this case that would have been the 300mm, but it would have put us flat against the opposite wall and make for very uncomfortable operation. Instead, we decided to use the 240mm Apo Ronar which allowed us to be into the room a bit. We could do this with no worries about reflections as Ms. Baca, aware of the problems a varnished painting can create, postpones varnishing her paintings until AFTER they’ve been scanned!

The Scanning Process
Clients never really grasp how much equipment and how many people it takes us to scan a large painting on location until we show up on the day in several vehicles. Fortunately, we’ve been through this with UCLA before, so they know the drill. They provided two people to prep the location and manage the large painting and we provided an assistant cameraman and scan-back operator (my wife, Jean) plus me and the artist herself, who graciously let us put tape all over every surface of her home while providing us with really good coffee.

After unloading the equipment, the “crew” (my assistant and the two men from SPARC) used a combination of studio black curtains and black foundation wrap to block sunlight from all the windows and skylights. It is not necessary to block 100% of the outdoor light from the room, so long as what there is remains constant all day, so they placed a large silk butterfly frame over the large windows visible at the end of the loft.

 
 

While this was going on, my wife and I began laying out the shot by measuring and taping lines and marks all over the place. The first line on the floor is the lens axis placed (in this case) mid-way in the room, perpendicular to the plane of the painting. I use a 4’ drywall square to do this because it’s big and yellow and hard to lose. Next, we placed 2 lines at 45° to the lens axis intersection with the wall as a guide for the lighting placement. Finally, we marked the image frame top, center, and bottom on the wall based on our computations for the tile size which is 67” top to bottom. You’ll notice that the left and right tape frame lines are missing. This is because we don’t need any more than the small piece of blue tape under the corner of the painting to indicate the left frame line (stop) as the painting travels through the frame, in each of the 5 tile positions.

Having taped out the frame on the wall, we placed the camera where it needed to be, matching the lens height to the same height as the center cross on the wall. Once this was roughed in, we began the black art of fine-tuning our framing and x/y/z perpendicularity using the Zig Align cascading mirror system, a tape measure, and an 8’ level. Our particular problem with this setup is that we based our Zig-alignment on the wall (which was plumb), not the painting, as the painting was hung on hooks from the trolleys on the 30’ steel track. We determined that a wooden shim 1” thick placed under the bottom edge of the painting at each stop could be used to keep the painting parallel to the wall at each position.

layout
 
 
setup Lighting the Painting
The goal here is to create perfectly even light across our frame which we calculated to be 67”h x 50”w. To do this, however, we need to light a slightly larger area to this same precision. This is why our diffusion panels are 72” high. Everyone has their own formula for lighting. I use an incident meter with a flat disc to get within 0.1 f-stop, then use the Better Light software eyedropper as a “spot-meter” to see if further tweaking is necessary. This would also reveal transmission fall-off toward the corners, which is always zero with our lenses. It’s useful to note that we prefer quartz lights for large set-ups as the instruments have both spot/flood and barn doors to modify and shape the light. We also have tons of them.
 
 

Final Prep: the Better Light pre-flight check.
We begin by pre-scanning the Mac Beth color checker targets and Better Light focus card. From this, we can determine an exact exposure, neutralize the gray patches for the color temperature of the light (in this case 3350°K), and precisely focus the lens, all by using the Better Light ViewfinderTM software. This software also allows us to look at small areas at full resolution to double check focus and more importantly, look for any undesirable light “kicks”. Finding none, we scan the color chart which will be later used in creating an ICC color profile that is accurate for the existing light of that day. Note: we create a new profile for just about every shoot, as it is easy and there are always variables, like the line voltage on the Los Angeles west side is different from our suburban studio by almost 5 volts. This is not an issue so long as the voltage is constant, as our profile du jour accounts for this.

 At right: Ron and Jean Finley go through the 20 step Better Light Camera "Pre Flight" routine they've developed to avoid time-wasting, embarrassing (or disasterous) mistakes. Note the Better Light Camera back inserted in the camera in the foregroound.

betterlight preflight check
 
 

The scanning process
We have calibrated the painting’s left-right travel into 5 equal distances to generate the five lower tiles. We’ve chosen to shoot at an aperture of f16 which is optimum for the APO RONARS and allows us to shoot at an ISO setting of around 600 which is great. Each scan takes around 15 minutes, which is why my wife, who has both more patience and tenacity than anyone I know, manages these while I exchange pleasantries with our clients out of earshot.

After completing the first row of tiles, we had to confront rotating the painting and re-hanging it on the overhead trolleys. The UCLA guys had tried this and knew it could be done. Unfortunately, this trial maneuver happened before we had lights and a camera in place, and couldn’t be accomplished with them there. So we had to carefully mark these positions, move them back and replace them after the painting had been re-hung. Other than adding some more gray hair to my head, this went smoothly.

 
 
inverted painting The "upper" row of tiles was most easily shot by simply inverting the painting. Note the industrial door sliding track on which the painting moved precisely into each new position, installed by Carlos Rogel of SPARC.
 
 

With the painting inverted, the five “upper” tiles were scanned smoothly. We checked all the tiles as well as we could with a portable computer, agreed they looked fine, and wrapped the shoot.

For any production managers reading this, this was an 8 hour day: 3 hour set-up, 4 hrs of scanning, and an hour wrap.

Post Production:
As always, the UCLA people needed the printing file asap, so we had some choices to make. First was which application would we use to assemble the ten tiles? WAIT! There’s another choice besides Photoshop?? Yes, and a good one: Panotools (PT/GUI). So we made the logical decision: do it both ways.

The Photoshop Approach
Using Photoshop as the app for assembling tiles is easy and logical, so long as every thing is shot perfectly. Create a new .psb doc (that’s the “big” version of a .psd), drag in the tiles, align them, create layer masks for each layer, blend them with a soft edge paint brush, flatten the file, and have a beer. But the reality is we're dealing with files way out on the edge of desktop computing's limits. I’m never quite sure what the maximum size is for Photoshop on my Mac this week. I’ve seen 300,000 pixels square, 5.4 GB file size, etc. All I know is that it’s more than a PC and I’ve never found it. I do know that when I get up in the 5 GB range, even with 20 GB of DRAM and 4 TB of drive space, things slow down a lot, as I suspect that we’ve begun to use virtual memory, even though we’ve set Photoshop to the max memory. (Why Photoshop doesn’t allocate memory dynamically is still a mystery to me) . I mention this because these files are huge and file size is a serious consideration for Photoshop.

Each of our ten tile scans was saved as an Adobe digital negative file (.dng) of around 700 mb which can either be converted to 8 bit or 16 bit .tif files. If we consider a Photoshop assembly with ten layers, the 16 bit approach would produce a working file of around 7 GB which would probably make even our computer gag. Besides, 16 bit is still not fully supported in Photoshop, so we converted the .dng to 8 bit .tiffs of 350 mb each, attached our ICC profile du jour, and rotated the inverted tiles. We were ready to put it together.

Rather than try to finesse all ten files at the same time, My plan was to do the lower row first, then flatten it to speed up the operation, then add the upper 5 shots individually. Because we had really generous overlaps horizontally, I had a lot of options in blending the layers. First, I align the five shots as accurately as I can, alternately varying the transparency and turning layers off and on to see if the image jumps. Next comes creating seamless transitions between the tiles. Generally, I begin by setting the upper layer’s transparency at 50% to start and drag over two layout lines to define the overlap. Next I add a layer mask to the upper layer and using a soft-edge brush of medium size (200 px?), I paint a random line through the mask that avoids areas of great detail. Setting the transparency back to 0%, I examine the mask transition at full magnification (100%) to see if there is any “ghosting” of the layers because of the soft edge brush or worse, any obvious differences in exposure. The ghosting problem can be corrected with a harder edge brush. Any minor exposure problem can be corrected with an adjustment layer (curves or levels) and a gradient mask. When I’m happy with the transition I’ve created, I paint out the rest of the mask and go to the next transition.

This all worked fine, and after a couple of hours of masking, I had finished the shot the scan assembly and sent off to UCLA for “comments”. So while I waited for the response, I turned to PT/GUI (Panotools Graphic User Interface) to see what it could do with our files. We prefer to use this program for Architectural panoramas as it allows us to auto-bracket the exposure of each panel with our SLR. A little-used feature of PT/GUI is that it can assemble tiles of the type we shot here (i.e. non-panoramic) with the press of a button through a mathematically arcane process of interpolation. The only caveat is that there be enough overlap to provide identifiable “control points” by which PT/GUI aligns the images, stretching and pushing them as necessary to make them line up. Our ten images had plenty of overlap and unique points for it to feast on so we loaded them in and pressed the “Create Panorama” button and took the dogs for a walk.

Were there concerns about file size or ram overhead?
PT/GUI has no file size limits other than those of your computer. It takes the available dram dynamically and uses scratch discs for the rest. The latest rage on PT/GUI forum is the HDR panorama. This is high dynamic range photography in which three brackets for each of 6 shots are mushed together in 96 bit space. Users write in to the Forum that their computer has been going for three days on a shot, and are told to be patient. But for our work, it can be ideal. Better Light pros readiing this probably cringe a the prospect of using a program like Panotools which has a bit of an image problem in purist photographer circles. In any event...

It took the PT/GUI about 15 minutes to create what it took me (and the computer) in Photoshop 2 hrs. I compared the results with the Photoshop version of La Danza and I can see no difference. I mention PT/GUI here because we’ve used it for more complex tile mosaics shot with smaller cameras, where we’ve had to move the lights and camera together as the original (e.g. tapestries) couldn’t be moved, sometimes assembling 50 tiles or more. Look out, Adobe!