Frederick Ferdinand Schafer Painting Catalog

Conventions: Image scanning and post-scan processing procedures


Scanning procedure

Prints are scanned on an Epson V350 scanner (since 2007), an HP 3500 scanner (2004-2007), or a Hewlett-Packard IIcx scanner (before 2004). The scanner is configured to "millions of colors" (24 bits per pixel) mode, using a resolution of 400 dpi, with neutral brightness and contrast settings and gamma of 2.2 (the IIcx scanner used a gamma of 1.0), and written in uncompressed TIFF format. These settings are intended to minimize image processing within the scanner and thereby capture as much information of the original print as possible. The IICX scanner creates output in the Macintosh RGB color space, and the other scanners create output in the sRGB color space. None of the scanners provides an ICC color profile with the image.

Transparencies and negatives are scanned by a Kodak (vendor 4220) FilmScanner 4045, and written in Kodak ProPhotoCD format. When the original is in 4x5 format, it is (usually) scanned at six levels; when in 35 mm format, it is scanned at 5 levels. For some kinds of original film material, the scanning system generates a custom ICC color profile that characterizes the scanner.

Minimal post-scan processing

As quickly as possible after acquisition, scanned images are placed on-line with minimal processing, using Photoshop 7.0 (version 5.5 from January 2000 through November 2002, version 3.0.5 before January 2000), as follows:
  1. Read the TIFF or PhotoCD image. If PhotoCD, read from a 1024 x 1536 PICT version of the image.
  2. If necessary, rotate the image slightly to make the sides of the picture approximately horizontal and vertical.
  3. Crop to the largest rectangle that fits inside the visible painting. This cropping omits the frame, and it may also omit corner wedges of the painting itself, if it is not square or the film plane was not parallel to the painting plane.
  4. If necessary, rotate 90 degrees to make the picture upright.
  5. Using Lab color mode, adjust levels to linearly stretch the brightness (luminosity) range of the image to span the full brightness range of the display. (This adjustment produces the effect of shining a bright light on the painting.)
  6. If the image was scanned from a print using gamma of 1.0, adjust gamma to 2.2.
  7. If there is an obvious color cast (e.g., green sky, red snow, etc.), roughly adjust the color balance. (See color adjustment, below.)
  8. Reduce size to 5-inch maximum dimension at 100 dpi
  9. Sharpen with Unsharp Mask, using parameters of 50% enhancement, 1 pixel averaging radius, and 0 difference threshold.
  10. Save inline image in jpeg format with moderate compression.
  11. Reduce size to 1-inch minimum dimension at 100 dpi and save thumbnail image in jpeg format with moderate compression.
The resulting image is useful for identification and comparison, but it is often not a very good representation of the original painting.

Standard post-scan processing

When time is available, images are more carefully processed. The primary additional step is to remove any keystone distortion and to adjust the ratio of the dimensions to match the original. In addition, in the case of PhotoCD's, work begins with the highest available resolution PCD image, acquired with the help of the appropriate ICC color profile. Finally, if other available information suggests that the color is incorrect, more detailed color adjustment may be done. Here is the standard procedure, using the Photoshop version mentioned above.
  1. Read the TIFF or PhotoCD image. If starting from a PhotoCD, and a custom ICC color profile was created at the time of scanning, use that profile. Otherwise use an ICC color profile appropriate to the original film type, reading the highest resolution scan available into the Photoshop LAB colorspace.

    [Original image] [Image with keystone removed]
    Original imageImage with keystone removed
  2. Crop to the smallest rectangle that contains all of the corners of the painting.
  3. If necessary, rotate the image and distort it to remove any keystone, assumed to be caused by the plane of the film not being parallel to the plane of the painting. (Note that if the original painting is not perfectly rectangular, this step will distort it.)
  4. Recrop to produce a rectangular image of just the painting.
  5. If the image was originally scanned from a print, adjust gamma to 2.2.
  6. Adjust levels to linearly stretch the luminosity range of the image so that it spans the full brightness range of the display. (This adjustment produces the effect of shining a bright light on the painting, and it removes some color casts.) If the color of the resulting image appears to be significantly different from that of the original painting, adjust the color balance further. (See Color adjustment, below, for discussion.)
  7. If the quality of the image is good enough to warrant providing a large, downloadable version, save a temporary copy of it. and then reduce its size to 11-inch maximum dimension at 100 dpi, at the same time adjusting the height-to-width ratio to match that of the original painting. Sharpen it with Unsharp Mask, with starting parameters of 50% enhancement, 1 pixel averaging radius, and 0 difference threshold. Save the large image in jpeg format with moderate compression and ICC sRGB profile.
  8. Restore the saved temporary copy and reduce its size to 5-inch maximum dimension at 100 dpi, at the same time adjusting the height-to-width ratio to match that of the original painting. Sharpen with Unsharp Mask, with starting parameters of 50% enhancement, 1 pixel averaging radius, and 0 difference threshold. Save inline image in jpeg format with moderate compression and ICC sRGB profile.
  9. Reduce size to 1-inch minimum dimension at 100 dpi and save thumbnail image in jpeg format with moderate compression and ICC sRGB profile.
If a detail image is created, it is obtained by cropping from the original scanned image and doing appropriate processing, rather than starting with a previously processed or reduced-size image. Signature and verso inscription details are level-adjusted as described above; they may also be enhanced by color adjustment and more aggressive sharpening to make the signature stand out better from the background.

Any deviations from this procedure are mentioned in the Adjustments section on the image information page.

Color adjustment

Of the 700-odd painting photographs collected so far, perhaps 5 provide an excellent representation of the color of the original painting. That is, the painting was correctly lit, the film was appropriate for the lighting, dyes in the negative or print have not faded with time, the scanner did not introduce any casts, a color management system followed the image through to post-scan processing, and nothing else went wrong along the way.

The others don't look like the original. To produce an image with colors closer to the original, many of the images need at least some color adjustment. This adjustment is by nature subjective, so one should consider the colors seen in these images as representative, not as authoritative. (The only consolation here is that the unretouched images were even less representative.)

Adjustment is done using a gamma 2.2 monitor with 6500 K white point that is calibrated by eye to make a scanned image of the Kodak Q60 color calibration target look as similar to the original as possible. Color balance is done whenever information is available that suggests that the image does not represent the actual appearance of the painting. In the case of older, faded transparencies and mislit photographs that show, e.g., yellow sky, pink snow, or green boulders, the need is often obvious.

[Original image] [image after color adjustment]
Original imageRed, green, and blue linearly stretched

In most cases, color and brightness adjustment is accomplished by simply stretching the Lab color mode luminosity to cover the full range from 0 to 100%. If this leaves a significant cast, the stretching is undone and the next thing tried is to switch to RGB mode and separately stretch the Red, Green, and Blue luminosities so that each covers a 100% range. If, as in the example at the left, there is at least one region in the painting that is completely black (one of the trees in the canyon below) and one region that is completely white (a rock on the trail at the lower left), this procedure automatically eliminates most color casts introduced since the painting left the easel--errors in scanning, errors in the photograph printing shop, shortcomings in the film, fading of film dyes, color introduced by the lighting, and even color casts contributed by aging varnish. When this technique is applied to an image of a painting that is to be cleaned it often produces a result that predicts what the painting will look like after cleaning.

This technique does not always produce a satisfactory result, particularly when a painting has no completely white region. In such cases, color adjustment is done by eye. Whichever method is used, the result probably does not exactly match the appearance of the painting, but it is usually much closer than the unadjusted image. Two important principles are that color balance adjustments are applied uniformly to the entire image, and are linear over the brightness range.



Title list
Other lists
Help About the Artist Copyright Notice Schafer Catalog Home Page
Comments, corrections, or questions: Saltzer@mit.edu
Feb 19, 2017, 09:57 MST