Dufaycolor - scanning, restoring, archiving

[e2b]

I recently came across a number of Dufaycolor slides, and because I could not find anything much about scanning and restoring slides made with this short-lived process, I thought might be useful to share my experience (in a rather long post ).

The Dufaycolor process was available from the mid-1930s until the early 1950s. A search on 'Dufaycolor' turns up quite a bit of information about the process and how the films were made and how they deteriorate (for example, http://www.aiccm.org.au/docs/Bulleti...2008_Vol31.pdf). However, I didn’t come across anything that was of any particular help in scanning Dufaycolor slides, so I made some tests myself, and have summarized my experience under five headings: the structure of Dufaycolor slides, appearance, scanning, image restoration and recovery, and an idea for archiving as a synthetic RAW digital image.

1. The structure of Dufaycolor slides

In the Dufaycolor process, B&W emulsion is exposed through the base and though a filter (‘reseau’) consisting of a mosaic of red stripes and green and blue rectangles mechanically printed on the film base, under the emulsion. The film is reversal processed and projected back through the reseau to produce an image consisting of tiny RGB rectangular elements, each more or less darkened by the remaining silver. Color formation is therefore additive. Before I had worked out that I was dealing with a Dufaycolor slide, my first impression when I scanned one was that it was a photographic copy of a screen printed picture from a book or magazine. However, the reseau is not really like a half-tone screen; it is more like a Bayer screen used in digital imaging.

The reseau can be seen in the following micrograph (from http://sasesearch.brighton.ac.uk/glossary/dufay.php ). (The red lines are actually angled at 67 degrees to the edge of the film roll.) The reseau was changed over the years; some illustrations show blue and green squares (rather than rectangles), and the spacing of the red lines was eventually reduced to 533 lines per inch.

Reseau micro.jpg

2. Appearance:

I had not come across Dufaycolor previously, the slides were not identified as being Dufaycolor, and I did not immediately recognize what they were. All the slides were 2-1/4 square, in plain cardboard 2-3/4 inch slip-in mounts, with no markings. The edges of the film roll had been cut off to fit in the mounts.

To the unaided eye, the slides were very dark, rather brown, and hardly look like color slides at all. They were quite dirty and scratched, and when after initial scanning (and with the owner's permission) I cleaned them with PEC-12, a lot of brown material came off the emulsion side. Most showed some silvering of the emulsion.

Although the reseau pattern is the defining feature of Dufaycolor, it not visible to the unaided eye. It is visible through a magnifying glass, and when the slide is projected, and it tends to be very prominent on a scanned image due to moiré fringing with the scanner and the display. Depending on the image, a potentially obvious naked-eye identifying feature is that (unlike conventional slides) Dufaycolor slides are viewed from the emulsion side.

When I scanned the slides it was evident from relatively light areas such as sky, that the dyes in the reseau had faded, and that fading was less in the edges that had been covered by cardboard mount. The mode of fading is quite distinctive, because while the dyes in the reseau fade, the silver in the emulsion does not fade, so that dark areas stay dark.

3. Scanning:

After some tests with an Epson V700 scanner, I concluded that 3200 dpi resolution was satisfactory for the purpose of image restoration. Scanning at 1600 dpi lost detail, while scanning at 6400 dpi did not show significant improvement. In order to minimize scratches, I wet-mounted the (unmounted) cleaned slides to the under-side of a Doug Fisher Variable Height Mounting Station, and to minimize moiré, I angled the slide 23 degrees, so that the pattern would be 45 degrees to the scan direction (this is not essential). I scanned in 16 bit RGB. There is no point in scanning with infra-red, since this is absorbed by both the silver image and the dyes in the reseau. This means that there is no useful image information in the IR channel. Neither is IR cleaning possible.

The red, green and blue dyes had faded differentially, and because the exposures cannot be set individually on the V700, I left a space in the mask to allow VueScan to base the exposure on a clear patch, and then locked the exposure by trial and error to the maximum allowable for red in the image area (typically in the range 8x to 16x), and selected 16-pass multi-scanning. (The Epson V700 scanner is not mechanically or thermally stable enough to combine separate exposures for R, G and B at 3200 dpi.)


4. Image restoration:

The only point to note in restoring color is the need to select an appropriate color sampling size (typically 31 x31) to give an average of the red, green and blue elements in the reseau. The main challenge in image restoration is minimization of the visibility of the reseau, without losing too much detail.

In attempting to minimize the visibility of the reseau, I tried just about every method of de-screening that I could find. PhotoShop Box Blur worked reasonably well, but best results were with using the Descreen 5.0 Plug-in. In order to smooth the appearance of areas of uniform color, I found it useful (in some cases) to apply Lens Blur, and Smart Sharpen (through masks) after de-screening. The owner, who had not been at all impressed with Dufaycolor (and had only shot a few rolls), reckoned that the restored photos looked better than the original slides ever had!

This picture of a PanAm plane turned out to be quite useful as a test since it has large areas painted white, a sharp transition to the dark interior, and lettering in various sizes. Just behind the door it is possible read the name “STRATO CLIPPER ECLIPSE”. This is at the very limit of resolution set by the reseau, as the letters are poorly formed and are red, green or blue, rather than the black. This odd effect arises because each letter is predominantly linked to a particular element in the reseau – needless to say, it disappears when the image is converted to B&W.

Copyright 1952 W J Murrell: PanAm.jpg


5. An idea for archiving as a synthetic RAW digital image :

The poor stability of Dufaycolor dyes was acknowledged while the process was in use, and these particular slides had faded significantly. Clearly, any important set of Dufaycolor slides should be copied in a suitable archival format. But what should be copied, and how?

It would be a very difficult task to preserve the particular ‘look’ of a Dufaycolor slide as there is no way of copying the unique structure of a Dufaycolor slide, with its built-in (faded) filter reseau. A photographic copy on color film could, however, record the current state of the image.

The alternative of a direct digital approach to preserving the structure of the reseau would also be a challenge. From the micrograph above, I estimate the dimensions of the green and blue rectangles to be 0.029 mm (0.0011 in) by 0.024 mm (0.00095 in), while the red lines are 0.019 mm (0.0007 in) wide. Scans with the Epson V700 at 6400 dpi show the reseau cells distinctly, though not accurately. However, by ‘eyeball’ estimate, the color value of each pixel is an average of an area with a diameter of around 3 to 4 pixels. In other words, at spatial resolution of 6400 dpi, the optical resolution of the V700 is only around 2000 dpi. Even scanning at true optical resolution of 8000 dpi would barely be sufficient to reproduce the details of the reseau (and the resulting file would be extremely large).

The task would be simpler if the objective were only to preserve the scene that was photographed. A digital restoration as described above would be sufficient for most purposes, although it is not entirely satisfactory, as the multi-colored appearance of the letters in the name demonstrates. However, there might be better way: this is to consider the reseau as being equivalent to a Bayer screen (albeit an unusual one with unequal elements, and angled at 67 degrees). In this approach all that would have to be recorded is the representative RGB intensity for each cell of the reseau. Then by an equivalent of Bayer de-mosaicing – interpolating values for the other two colors at each cell – the image could be recovered.

The idea that the reseau consists of cells comes from considering the red lines in the reseau to be rectangles placed end-to-end, adjacent to the green and blue rectangles. Thus, the basic repeating unit in the reseau can be considered as a square with sides 0.048 mm, consisting of one green, one blue element, and two (adjacent) red elements. In the micrograph shown above the red cells would have dimensions 0.019 mm by 0.024 mm (0.0007 in by 0.000095 in).

I have made two rough tests to confirm that such an approach might work.

The first test was to see what sort of scan would be required to get a representative value for each element. I used the Epson V700 set at 6400 dpi to scan an area of uniform white in an image, for which there should be no color variation within each cell of the reseau. This revealed the cells but, because the optical resolution is only around 2000 dpi, only the few central pixels in each green and blue cell can truly represent the color of the cell– the other pixels are averages of the colors in the adjacent cells. Because the red cells are so small, this resolution is not quite good enough to give a true representative value – all pixels in the red cells are contaminated by the adjacent green and blue cells.

The second test was to see if each cell can be regarded as having a single RGB value. The scanned section shown above includes a sharp boundary between the white fuselage and the black doorway. Looking at the line of transition between light and dark, the 6400 dpi Epson V700 scan shows that there is a visible change in tone within those green and blue cells where the transition occurs, but that the red cells are too small to show any change. However, while this tonal variation is real, my testing convinced me that it is not a significant factor in the resolution of the image, which fundamentally is set by the size of the reseau elements.

For this test I took a small (109 x 78 pixel) section of the 6400 dpi scan showing part of the name of the plane. In this sample it is just possible to read the letters “TRA”, when displayed at 100% or lower magnification. Then, working at 800% magnification, I painted each individual reseau element to be the same color as its central pixel – I did this quite roughly, by hand. The resulting image looks very crude and blocky, but when viewed at 100% magnification, the visibility of the letters is essentially unchanged.

As scanned: Name.jpg Hand-painted to color of central pixel: Name hand-painted.jpg


This means that for practical purposes tonal variation within each of the four cells in the basic reseau unit can be ignored. For me, this was sufficient confirmation that a Dufaycolor slide can be thought of as being a like a RAW digital image taken through the equivalent of Bayer screen, which means that it could be reproduced from just one RGB value per cell on the reseau/screen.

On this basis, 2-1/4 inch Dufaycolor slides are like 5 - 6 Mp digital camera images. If anyone ever found it worthwhile to do so (maybe for restoring one of the few movies made with Dufaycolor), it should be possible to write software to locate the reseau elements on the scanned image automatically, record the RGB data from the central pixel in each cell, and use them for de-mosaicing the raw data from the reseau/screen. A scan at a true optical resolution of 8000 dpi would be sufficient for this, and the large file produced would just be an intermediate step in producing a much smaller ‘synthetic’ digital archival record.

I believe that the results of such de-mosaicing would produce better images than those obtained by direct scanning and blurring, as described above. Mathematical interpolation would avoid the inherent limitation of de-screening techniques for merging colors, as shown by the example (above) of the red, green and blue letters in the name of the airplane. Just a few thoughts …