The Drum Scan Gap

[Contents] [Previous] [Next]

The drum scan

In 1967, Josiah Thompson hired a professional Dallas photographer to copy the original Moorman Polaroid. The photographer returned to Thompson two 4 x 5 inch black and white negatives and 8 x 10 prints from the negatives. When the Moorman controversy arose recently, Thompson scanned one of the prints on a consumer-grade flatbed scanner and made the image available to the JFK research community.

Dr. Costella used a version of this image file for his gap analysis. As part of the analysis, he magnified the image 3 times larger and rotated it to compensate for both camera rotation and scanner rotation.

Both the Thompson original image and Dr. Costella's enlarged version contain compression artifacts, presumably from the source JPEG image file. The edges show signs of ringing, which is a byproduct of the compression inherent in JPEG, but also may be a sign of edge enhancement. Dr. Costella's processing of the image inevitably contributed its own artifacts.

Unsure of the effect of these artifacts on gap measurement, we decided to get as close to the Moorman original as we could. Josiah Thompson took his negatives to Octagon Digital in San Francisco to have one of them professionally scanned on a drum scanner. He had the 4 x 5 negative scanned at 2400 DPI with no edge enhancement, no tonal adjustments, and no other processing. The scanner generated a 110 MB, 8-bit grayscale image to CD as an uncompressed TIFF file.

Here is a comparison of a region of Dr. Costella's image (a) with the raw drum scan (b):

The Moorman Polaroid had significantly faded by February 1967. The original Thompson print and scan reflect this as does the new drum scan. Somewhere in the scanning and generation of the image (a), the tonal range was stretched to increase the contrast of the image. If done carefully, such contrast adjustments do not destroy information in the image or contribute significant artifacts. Image (c) shows the effects of PhotoShop's Levels adjustment on the drum scan image (b).

The drum scan of the negative appears noisier than the flatbed scan from the print. A drum scanner can record far more detail than a flatbed scanner, typically up to at least 8000 DPI. At 2400 DPI. the scanner begins to resolve the individual grains in the negative. The photographic enlarging and printing process filters the grain so that it does not appear in the final print. This is akin to turning down the treble on a high-resolution digital recording of an old tape recording to reduce the tape hiss. Some "blur" filters used in image processing achieve a similar effect by filtering high-frequency detail. Image (d) shows the effect of PhotoShop's Gaussian Blur filter on image (c).

 

The drum scan gap measurement

After distributing copies of the drum scan, we went to work on measuring the gap size between the lower edge of the window and the upper edge of the pedestal using Dr. Costella's edge location method. Joe Durnavich performed the new gap measurement with guidance and assistance from the rest of the group, especially David Wimp.

Dr. Costella's method is not well suited to the raw drum scan because of its low contrast and high grain detail. I adjusted the drum scan's levels and filtered the result with Gaussian Blur, the same procedure as used to generate image (d) above. This produced an image more comparable to Dr. Costella's, but without the compression and other artifacts.

Here is the result of the drum scan gap analysis with the edge lines and gap sizes marked. The gap size here is 23.97 pixels, but note that the drum scan image is larger than the one Dr. Costella used. On equal terms, the new gap measurement is 3.4 times larger than Costella's. If the edge lines drawn here don't look like they are positioned on the actual window and pedestal edges, keep in mind that due to the motion blur in the Moorman photo, a correct edge line position falls between the transition from the light area to the dark one.

I performed the following processing steps in PhotoShop 5.0 LE to the original raw drum scan image:

  • Cropped the image to include just the area of the upper pedestal and 3 of the windows.
  • Adjusted Levels with the black point set to 105 and the white point set to 217. Checks with the Threshold function verified that no tonal detail was lost due to clipping.
  • Applied the Gaussian Blur filter with a radius of 4.0 pixels.
  • Saved the result in .BMP format, 8-bit grayscale.
  • I chose the Gaussian Blur filter because as a low-pass filter, it does not introduce any noticeable ringing artifacts. Tests with edges made in Microsoft Paint verified that the Gaussian Blur filter does not change the location of the edge using Dr. Costella's edge location method.

    To avoid further processing, I did not rotate the image to make vertical lines in the pergola structure vertical in the image. Instead, I also measured along the right window edges to locate the local vertical and factored that into the final calculations. The resolution of the drum scan was high enough that no enlarging was necessary.

    The Moorman Polaroid is marred in places by photographic defects. Someone who handled the Polaroid shortly after it was taken left a fingerprint on the right side, which has been eroding the emulsion in several places ever since. The fingerprint extends into the pedestal and window region. Such defects prevent the measurement of edge locations at particular points. Statistically, these points are usually readily identifiable in the data because of their wide variance. Visually, an edge is not discernible. I excluded these points from the final results.

    The data sets include over 500 edge positions. The numbers of edge positions by feature are:

    (V) Vertical 195 points
    (WT) Window, top edge 170 points
    (W) Window, bottom edge 150 points
    (P) Pedestal 70 points

    Most of the window edges are measurable. As Dr. Costella found with the pedestal, only a few areas are suitable candidates for measurement. The selected points include 19 near the left side of the pedestal, about midway between Zapruder's right foot and the pedestal's left edge, and 51 points by the right edge, where a dark object rests on the pedestal, presumably a purse.

    When graphed as a set of points, the edge positions generally lie along straight lines representing the edges. Microsoft Excel includes a statistical function, LINEST, that finds the line that best fits a set of data points. The LINEST function generated the slope and intercept values from the sets of edge positions, giving these equations that describe each line:

    	(V)  y =  37.16339x  - 8864.7777
    
    	(W)  y =  -0.060135x -  449.1539
    
    	(P)  y =   0.001266x -  486.9938
    

    The intersections of these lines are the following (x, y) points. The distance between these two points is the size of the gap:

    Intersection of V and W = (226.083, -462.749)

    Intersection of V and P = (225.439, -486.708)

    Distance between them: 23.97 pixels.

    To verify that my edge measurement method led to consistent results, I also measured the top window edge, which resulted in an edge line with an only slightly different slope than the lower window edge.

    Intersection of V and WT = (228.662, -366.908)

    Distance between that and lower window corner: 95.88 pixels

    Gary Mack measured the actual window height and found it to be 11.0 inches.

     

    The gap size and the camera position

    Dr. Costella calculated the gap as being 2.9 pixels in his "zoomed" image. This image is enlarged 3 times from the original 1364 pixel-wide version. The corresponding image area of the drum scan is 9890 pixels wide, making the drum scan 2.42 times larger than the Costella image.

    The drum scan gap of 23.97 pixels is then:

    	(23.97 pixels * ((1364 pixels * 3) / 9890 pixels)) / 2.9 pixels = 3.42
    

    or 3.42 times larger than the Costella gap.

    Dr. Costella calculated that each pixel of his gap equates to 0.01576 feet at the pedestal position. The 2.9 pixel gap translates to 0.55 inches at the pedestal position.

    To find the gap size at the camera position, one must divide the distance from the pergola wall to the camera by the distance from the wall to the pedestal and multiply the result by the gap size at the pedestal:

    	(132 ft / 37 ft) * 0.55 inches = 1.96 inches
    

    This would mean that if Mary lowered the camera 1.96 inches, the gap size would be zero. The top edge of the pedestal would be even with the lower window edge.

    (Dr. Costella used a factor of 4 here in reporting his results, making the gap at the camera position 0.55 times 4 or 2.2 inches.)

    The drum scan gap, being 3.42 times larger than the Costella gap, translates to 1.88 inches at the pedestal, and 6.7 inches at the camera position.

    These new measurements produce a camera position that is consistent with the Moorman location on the grass seen in the the Zapruder, Nix, Muchmore, and Bronson films.


    [Contents] [Previous] [Next]