Tigertail Museum Development and Restoration

Background of the
Development of the
Tigertail Virtual Museum

The Tigertail Virtual Museum has always been dedicated to the goal of restored pictures. But for at least the first five years the goal of restoration has migrated in meaning.

Year zero, 1998, the year before the first version of the website was published, I just collected art from the fine-art newsgroups. These groups scan and distribute art to their group of appreciative admirers. Most scans that proved useful to our museum came from the group alt.binaries.pictures.fine-art. Usually the scans were of moderate size (800 pixels high) and usually quite good, but occasionally one saw a really bad scan, and most reflected a yellow tinge that is almost ubiquitous in art books.

Several times I pointed out that such scans were lacking in accuracy, both to the picture in the museum and to original intent of the artist. I provided modified versions of the scan, to show what I thought the art should really look like. I hoped to start a discussion of what the art should really look like, and the causes of the changes in color. Instead the group was horrified. No one was to change a posted scan—that was sacrilege. The self-appointed leaders, defined and posted a set of rules governing the group that prohibited reposting of modified images.

Obviously I was not going to get a discussion going, nor it seemed was anyone interested in the problem of how the art looked. I wrote a program that automated collecting images from the group and decided to just dump inadequate scans.

Several months into this project I began to try and identify and organize the thousand or so images I had collected. I looked at what the program had collected and discovered a great deal of porn and all kinds of miscellaneous odd images. This was entertaining but the first task was to sift the wheat from the chaff.

The second big surprise was that that the file names associated with the image files tended not to reflect the artist and name of the picture. One might get a name like "bel01.jpg" or "1952.jpg". Both of which give little clue

as to what the picture really might be. Sometimes a more descriptive title might emerge: "mich_sistine1.jpg". So I set out with three friends to try and identify and organize the pictures.

We built a historical tree with directories with artists names in directories in the tree and began to rename the pictures using the pattern:

artistname_picture_name.date.jpg.

If we could not find a date for a picture we usually just left the date off.

When inadequate or really bad scans were found they identified the picture and sent it to me to try and fix. I found I could usually edit the picture with Photoshop get a satisfactory image out of these scans; so then we would put it in the proper spot and continue sorting. Every night more images from the newsgroup would pour in and more sorting and identification would need to be done.

Sometime around the beginning of 1999 I approached the board of Tigertail Associates about creating a website with the material I had collected. While not enthusiastic they felt that if I did the work as a volunteer, they would fund it to the extent possible. We went looking for someone who would volunteer to host the site for free on the Internet. Zenith Internet Services, an ISP, in Los Angeles was willing to host the site and we had our home on the Internet.

We discussed the copyright implications of republishing the pictures. I argued that as a non-profit organization dedicated to furthering arts education and promoting technology and the arts, Tigertail was carrying out its non-profit mandate by creating an online museum. And copyright law allowed for non-profit educational use as "fair use". I also argued that the pre-20th century work was in the public domain. (Although most museums, publishers, and image libraries don't seem to think so; however last time they tested it in court—Bridgeman vs. Corel, 1999/2000—they found that at least that court didn't agree with them.) The Tigertail board made the decision to try and build a virtual web museum, with example images covering western art from cave art to the end of the 20th century.

As we began to seriously work on this project I discovered that both disk space and computing power were limiting what we could do. I approached, Steven Kirsh, for some funding for new equipment and his foundation gave $20,000 to Tigertail Associates to buy equipment and to help keep us going. We got our first reasonable sized machine with large fast hard drives, 40 gig if I remember properly, and began to work seriously at building the website.

Arcimboldo: Water
[Before Restoration]

Arcimboldo: Water
[After Virtual Restoration]

The first version of the website came on-line in January of 1999. One thing we learned immediately was that practically nobody entered the museum through its front doors. Most got to the site by a link from somewhere else that pointed an artist's page. Our little group sat down and discussed the result and set out to redesign the site. Thus the top banner on every page was redesigned to help redirect new arrivals to other places in the museum.

Using a digital 3D modeling program, 3D Studio Max, we built a three dimensional model of the museum so that we could show virtual scenes of the museum and its environment.

We also changed direction a bit, instead of just organizing and identifying what the news groups were broadcasting we began to fill in the missing areas of western art that the newsgroups tended to not be interested in. One of these areas was a major modern artist, that we felt should be represented even if the news group members wanted to ignore him.

So we borrowed an old HP scanner from Zenith Internet Services and entered our first scans into the museum. The redesigned site took over a year to build and version two of the site came on-line in February of 2000. At this point the majority of scans were still coming from the newsgroups.

This new version featured three different size images for each picture with the default size being 750 pixels in the longest dimension. This required making a standard archive image for each picture in the largest sizes, which is 1000 pixels in the longest dimension, stored in a non-lossy format. These reference images tended to be pretty big, about 2 to 4 meg per image. We developed a program which would process the reference image and produce four images: a thumbnail, large, medium, and small jpeg version of the image in the proper place in the website. Jpeg pictures use lossy compression, so they introduce noise at high magnifications, but they reduce sizes of large images to 100/200 Kilo bytes, which is suitable for distribution on the network.

We standardized on the image standards employed by the PC, rather than the Mackintosh, because the vast majority of our visitors were PC users. There was some argument about duplicating the pictures for MAC users and delivering the appropriate picture depending on what computer the visitor was using; but we decided that route was too complex, that the majority of other art museums on the web were not providing that service, and that it would cost us a great deal of development time, to satisfy only a small percentage of our users, under 2%

In the course of 2000 and 2001 using our new technology and new website organization we attempted to fill in gaps in the early history of art, from the prehistoric, Egyptian, Greek, and Roman periods. Toward then end of that period we added a bit of Medieval art, although this remains a weak area of the museum. In the summer of 2001 The New York Times picked up on one of our more tongue-in-cheek restorations (see below) and did an article about the Tigertail website.

At the beginning of 2002 we realized that we had been focused to narrowly on current technology and that it would not be long when larger, cleaner scans would be necessary. So we purchased two state of the art scanners, one for scanning paper and the other for scanning slides.

We began experimenting with getting high resolution pictures from slides and printed books. This led to the discovery of many problems that exist in the creation of quality high resolution images. We have sought to solve these problems one by one. And our results continue to improve as our techniques and tools get better. But we have discovered that high-resolution images take much more time to restore properly. On the average it takes a trained artist from 2 to 3 hours per image to do an adequate job. However some images can take days, e.g., the Arcimboldo images took 8 to10 hours each (see above for an example).

At this point we began to store two reference images: a 1,000 pixel version for use in the web site, and a 4,000 pixel version, occupying about 35 to 50 meg per picture for future use. Source material is often not adequate to get a clean 4,000 pixel version of an image, so for these images we would save a 2,000 pixel reference image instead. Reference images are always stored using non-lossy compression to preserve each pixel properly.

We are now working to add material and replace older scans with high resolution scans of all the art. This will take a great deal of time and effort. Probably it will be five years before we have 90% of the images in high resolution format. Storing 35-50 meg per image requires very large amounts of reliable storage, and even DVDs are really cumbersome to keep backups of our archives. We do use DVDs for backups but it takes 25 DVDs and a week of work to make a full museum archive backup. We do this twice a year. However we have had disk failures that set us back months, so we are looking for a better solution for backups in general.

Leaning Tower of Pisa
[Before Restoration]
[Note that people are in serious danger
from this dangerous building]

Pisa Campanile
[After Restoration]
[People are safe and one can look at the architecture
without being anxious about engineering.]

Problems Getting
High Resolution Images
from Slides and Scans
of Illustrated Books.

Slides from slide libraries are usually pretty bad, while the resolution is usually adequate, the colors are usually terribly off. This comes from two problems: the source slide reproduction is not color consistent with the original slide and most slides in our local slide libraries are relatively old and are seriously faded themselves. However we found that with care we could bring back most of what was lost. This color change problem is covered later because it is similar to the problems encountered from getting pictures from books.

Printed illustrations are more of a problem. Difficulties here came from several directions.

Alignment of Color Plates

First it is common to have misaligned color plates. Photoshop will allow some correction, but several transformations are really missing if one wants to align color plates. So we have created programs to reprocess scanned books to realign misaligned color plates and a special smoothing routine which takes the printing dots and merges them together seamlessly, creating the highest resolution image from printed material that is possible. Often this is better resolution than one can get from commercial slides of the same pictures. See the examples below.

Color Correction

A second problem is metamberism in color printing inks. That is, what looks fine in white light turns yellowish in the light used to scan the picture. This is because the spectral qualities of the white light used in the scanner are different from the spectral qualities of the light you view the picture with. This must be compensated for by adjusting the colors manually in the computer to make a reasonable image.

However often the picture in the book looks a little yellowish itself. Sometimes a great deal yellowish. This can be caused by several different things. The most common with images of oil paintings is that the protective shellac covering the painting in the museum has yellowed with age.

Paintings must be cleaned of their shellac about once or twice a century or the shellac gets so discolored that one can barely see the picture. The major museums like the Louvre, the National Galleries in London and Washington DC, and the Uffizi, have much more art that their budgets will let them maintain; the result is that some important pictures from our art heritage tend to be tinted yellow.

Anyone who scans must (and does) correct for color. The question is what color should one try to achieve. If one tries to achieve what is in the book, you are not sure what you have accomplished. There is no guarantee that the photographer, printer, lithographer, and printers have done their jobs in any consistent way. Unless the publisher has held each and every step of the job to exacting standards the color will have drifted a long way from the original. For some reason, this direction tends to drift toward the yellow. Probably because this looks more reasonable to each person along the production line.

Alternatively the person can try and correct the scan for the color of the picture in a museum, under some ideal circumstances. This is difficult without access to the museum. And sometimes very difficult even with access to the picture. However, often the picture hasn't been cleaned recently and so the picture as it is in the museum doesn't represent the artist very well either.

What we do in the virtual museum is try to get to what the artist might have seen when he finished a picture. One has a reasonable chance of doing this because most pictures have an internal consistency in the colors of a picture. Usually garlic is white, grass is green, blacks are black, and white is white. Human skin usually looks like it should, not white and not horribly yellow. As one adjusts the color balance of a picture there usually comes a point where the picture "comes alive", and the restorer becomes sure that he is close to the right set of colors.

Sometimes this is impossible because the picture in the museum is so yellowed that the colors are completely washed out. For examples of this problem see the Arcimboldo example above and the Bellini picture example below. But we do the best we can.

Don't think that these errors of assigning the wrong color to something are peculiar to a virtual museum, the Louvre recently cleaned the Marriage at Cana. In the cleaning they discovered that one of the major figures had been over painted and changed from a green cloak to a red one. So they removed the red one to get back to the original. Unfortunately research later revealed that Veronese had repainted the figure himself. There was really nothing to be done—so the figure is now in a green cloak and not what Veronese wanted.

One can also be fooled by a picture that has been deliberately tinted for some artistic effect. But usually a little research on the picture and the artist yields the right information to inform the restorer of what needs to be done.

Close up of Printed Image
with misaligned printing plates.
[The magenta plate is to the right and below the other plates.
This causes a narrow halo of magenta
color to the right and below black lines in the image.]

Same Image
with plates properly aligned.
[The difference here is so small that you
need to examine the enlargements to really see what is going on.]

It can also be difficult or impossible to have the picture "come alive" if the colors in the picture and the colors from a CRT just don't come together the right way. This can happen when the restorer just doesn't hit the right point, either from inexperience or because of something innate in the picture itself. Paint produces colors by subtracting spectra from the spectra of light falling on it. CRTs produce color by lighting up phosphors of color which your eye merges together into the impression of some color. So some paintings just seem to resist being mimicked well on a CRT.

Dynamic Range in an Image

The last major problem is one of dynamic range of the image. The human eye can perceive differences in the very dark and the highly luminous. The eye is far more sensitive than photographic film, which in turn is more varied than photographic paper on which it can be reproduced, which in turn is more sensitive that what can be duplicated on a printed page, which is turn more varied than what a CRT can shine into your eye. So a painting's “art” can be hidden from even a photograph if the artist uses the full range of what the human eye is capable of. The trick in photography is to try and map the dynamic range of the subject to the photograph. The job of the printer is similar—as is the job of the person who scans a picture for the web.

If one doesn't take this responsibility seriously then the resulting representation on the web will usually be too contrasty and one will not be able to see what the picture is about. A good example of this is the Bellini picture in the next section from the US National Gallery of Art website. This is probably an accurate representation of the photograph of the picture after its museum restoration, but it conveys little of the picture to anyone who really wants to see it on the web.

The cause of the contrast problem may be that the web designers working for the National Gallery of Art may be using a MAC and be totally unaware of what others are seeing. It is better to err in the other direction, that is work on a PC and let the MAC people see the result, which tends to lower the contrast of an image. Another explanation is that the web desigeners were given a scan and they just placed it on the website, as is, because they felt they had no right to change it.

So in this museum we try to seriously map the dynamic range so you can appreciate the picture without doing damage to the original intent of the artist. We use the PC standard gamma, rather than the MAC one.

Artifacts Change as they Age

As a picture ages not only does paint fade, but worms can eat holes in a picture, paint can crack, sculpture can be dropped, or broken on purpose. If these changes interfere with seeing a picture as it was—then we try to repair the changes. If many fine cracks occur, then it is usually not necessary to fix anything because the human eye will ignore them anyway. However if major cracks disrupt viewing then we try and repair them (see the example below.)

Some times our virtual restoration goes even further, sculpture may get noses, cracks in paint get removed, and in general we try to clean up the image so one can see what the artist intended.

Greek and Roman sculpture was painted gaudy colors. Time has removed most of the paint. The paint can still be found in some of the corners of a sculpture. There are some examples in the Greek and Roman section that have been repainted so you can get a feel for the originals, but in general these older sculptures are left plain. Usually however we replace noses and other small parts that because they are missing distract one from seeing the object (see the Kritios Boy example below).

Sometimes, since we are quite enthusiastic, we even go further that the limited changes described above, and restore a work to what we think the artist intended, but never achieved. For example the Leaning Tower of Pisa, does not lean in this museum, even though it started to lean as the workmen built it. The tower viewed this way has interesting things to tell us about its design. One can no long marvel at its standing but you have to look at its design and it's relation to the buildings around it. That is you have to look at it as art. See the before and after restoration sketches above. This was sufficently amusing that the New York Times featured it in the arts section.

Most virtual art museums (as well as the real ones) don't tell you what they try to accomplish when they publish a web image of some art, perhaps they haven't thought out the issues, or they just feel that people don't really care what they do. We feel we should be clear about the changes we make and why we make them.

There are many other art resources on the net, including other web museums. Some can be reached by clicking here.

Restorartion for
Cracking and Color

Color Correction —
An Extended Example

Color restoration from a picture printed from a photo taken by a museum of a picture they have not maintained for over a century is a challenging task. The shellac used to protect the paint yellows with age, about every 40 years or so it needs to be removed and the painting reshellaced. By the time 100 years have past much of the life in the painting underneath is difficult to see. The process of cleaning the painting is costly and somewhat dangerous, since every time you remove the shellac there is a good chance some of the painting will go with it. Many artists complain that American restorers and some European ones tend to over-clean and thus destroy a picture rather than revive it. On the other hand if you leave it unattended there isn't much of the picture left either.

Over time pictures develop "problems"; the media on which they are painted can crack, crumble, or disintegrate. The paint itself will crack with age; lead-white paint is particularly prone to cracking. The usual museum restoration technique is to coat the painting with a layer of shellac, and then paint over the problems. Then shellac over the painted restoration. Thus what you see is the restorer's version of the painting. The amount of repainting depends on the restorer, the damage to the underlying painting, and the instructions of the curator (or owner) responsible for the painting.

When we digitally restore a painting the first thing we try to get rid of is the orange/yellow from the old shellac. However when the shellac is pretty old the yellow is so dominant that it colors everything else. Blue tends to become green, light blue can actually become yellow, gold becomes a darker gold, and white becomes yellow, red tends to become brown and so on. Worse some colors begin to merge together, a yellow robe, and a white one can begin to look the same. If you just remove some yellow then gold looks purple and white still looks a lighter yellow. If you remove all the yellow then the picture is not going to look right at all. In these cases one needs to make educated guesses about the real colors involved, and do some colorization to try to get back to where the picture started. A good example of this is shown in the Arcimboldo example above.

In 1987 through 1990, The National Gallery of Art in Washington, DC, cleaned a Giovanni Bellini picture, Feast of the Gods. To great fanfare this was shown to the world in 1991. The picture provides a way of seeing how much shellac changes colors. Below is a before and after picture. The only change made to the images was to enlarge them to 1000 pixels so that Tigertail members see a detailed comparison.

When we started our Virtual Museum in 1998, this painting was represented by a scan that was not a complete image of the painting; unfortunately it was from a post-restoration photo. So it is not interesting to compare it with the above two pictures. The question remains, what would our restoration have looked like if we had gotten a scan like the one from the Mark Harden Archive website? I decided to test this by giving this Harden scan as a test restoration to an art graduate student that does restoration for us from time to time. I didn't do it myself because I had seen the museum restoration. I asked him not to do the usual background research, because then he would find the museum restored picture and the results would be influenced by it. I told him he only had four hours to work on the restoration.

He came back and told me it was a difficult picture to work on because so many colors have disappeared into the yellow-brown in the original scan. I showed him the picture from the museum website. He was shocked at how bad the museum reproduction was and fascinated by the colors revealed by the actual restoration.

I was also fascinated by the comparison between his restored picture and the museum restoration, even if the details differed significantly—the all over feel of the two restorations is pretty much the same. One place that is significantly different is in the upper left side. Our graduate student could not distinguish between the mountain

and the sky clearly, so the coloration in that area of our restoration is similar, whereas the museum could really see the color of the paint, so they were not about to make that mistake. Clearly he also guessed badly on the color of the clothes the Gods are wearing. But I don't think this significantly changes the feel of the picture.

One comment that the student made was that he felt that more blue and yellow were needed in the picture and if he had more time he would have found appropriate places to add them.

One major difference between our images and the others is that the contrast on both the Mark Harden image and the one from the Museum is very high. This could be caused by using a Macintosh computer to process the images for the web rather than a standard PC. Another explanation is that the results are just as the scanner delivered them, with neither site taking time to try to correct the contrast. But on a PC it is difficult to understand the picture because of the high contrast.

The reason that one can restore something from as badly kept a copy as the pre-restoration US National Gallery Bellini picture is because, in general, art before 1850 was made to look like reality, and reality is a long distance from the Harden scan, and anything that brought the colors back in line with reality looks and feels much better. Let it be said that Mark Harden's scan probably does look pretty close to the reality of what the US National Gallery was displaying in the museum before 1986.

Shellac is not the only problem, For many years white paint was based on lead to produce the white color. However lead paint tends to turn greenish if it has not been exposed to enough sun light. Museums tend to keep sun light away from paintings because it fades the colors. However to get a good white out of lead paint you need to expose it to light. When museums photograph paintings do they exposed them to sun light to get the lead paint to work right? Probably not, and the photographer probably isn't even aware of the problem, so he probably didn't filter for the color change either.

Few people who scan art images are willing nor competent to do the appropriate color matching. And that assumes the color in the book in sunlight, for example, is what represents the picture in the museum. Most printers certainly don't check colors against the original art; they depend on the lab that created the color printing plates to have done a good job. The lab depends on the photographer having done a good job and so on. So what you see is what you see; it's nobody's responsibility they all just did their job.

It is a wonder that the art images that are commonly available come close to the ideal at all, and the truth is, in a book or on the web we often see something far from what the museum has in it, even in a museum book shop. Even worse what you see in a real museum when you look at the picture the artist created may well be far from what the artist saw when he painted the picture.

It should probably be said that many people like the old aged look of master paintings. There is something in us that automatically reverances the old. This website and what it trys to accomplish will not please people who's interest in art is driven by this feeling. But then there are plenty of other sites to visit if you want things to look old and yellow. We could create a website devoted to virtually aging art, where we take twentieth century art and artificially age it. Although that is an easier task than the one we set for ourselves.

We take care with our images, and we try to produce the best possible image, one that the artist would be proud of. We hope the pictures on this website look a little bit better than the average art book or other image website, and you enjoy what we have to show.

The Museum Director,
Robert Uzgalis.