Evaluating Claims of Zapruder Film Alteration

Part 2 – Line-Profile Analyses

by Craig Bouzarth

Background

Scan Versions. This article focuses on three scanned sources (see the appendix for further details).

Limits of Source Film Resolution. The following points summarize the relevant portions of the discussion in the Part 1 article:

Impact of Processing Adjustments. The 35mm copy purchased by Whitehead and Wilkinson is an internegative rather than a positive print, and any analysis must account for that fact. Converting the internegative to a positive after scanning requires correcting the orange mask inherent in color-negative film and inverting the colors. This conversion involves operator choices at several stages, including adjustments to contrast, gamma, hue, curves, levels, channel treatment, and sharpening. These adjustments can materially alter the appearance of the head wound, as the following figures show.

Figures 1 through 5 derive from the same 6K scan file; Figure 6 is presumably derived from a rendering of the same file. The figures nonetheless reflect different processing choices. Figure 6, a crop from the slide in Figure 22, shows a single color channel extracted from an RGB image, with a line drawn across it. The slide does not identify which channel was extracted.

Several figures below are drawn from Dr. David W. Mantik’s 2023 lecture at the Citizens Against Political Assassinations (CAPA) conference. During the lecture, Mantik plays a video in which Whitehead discusses the line-profile analyses with several postproduction colleagues. Figures 3, 4, 12, and 15 reproduce screen captures from the video shown during the lecture; Figure 5 is an author-adjusted version of Figure 3.

Figure 1. 6K scan (Source:[L2] Midnight Writer News)

Figure 2. Apparently different processing of the same scan as in Figure 1, producing a sharper image (Source:[L2] Midnight Writer News)

Figure 3. 6K scan version shown in a lecture (Source:[L3] 2023 CAPA)

Figure 4. Apparently different processing of the same 6K scan, shown in the same lecture as Figure 3 (Source:[L4] 2023 CAPA)

Figure 5. Version of Figure 3 lightened by the author, revealing more visible structure within the shaded region (Source:[L3] 2023 CAPA; image adjustment by the author2)

Figure 6. Crop of the unnamed analyst’s slide (displayed in Figure 22), showing a head shadow darker than those in Figures 1 through 5 (Source:[L2] Midnight Writer News)

Visible structure within the shadow is clearly apparent in Figures 3 and 5 but is barely apparent in Figures 4 and 6. Because Figure 6 presents a single extracted channel rather than a full-color rendering, it illustrates the effects of channel selection and processing. It does not provide a strict density comparison.

These six renderings show how widely the appearance of the head-wound region can vary with processing choices, channel selection, and possibly differences in scanning.

Changing Nature of Shadows. Figures 7 and 8 are repeated from the Part 1 article. Although unrelated to the assassination, they demonstrate how uncontrolled directional lighting and small head movements can substantially affect head shadows. In this example, more of the subject’s light-colored hair appears black after a slight shift in head position. Controlled professional lighting is used precisely because illumination in uncontrolled scenes can vary with small changes in position and angle.

Figure 7. Head shading of an audience member3 (Source:[L5] crops from a YouTube video)

Figure 8. The same head with different shading a split second later (Source: see Figure 7)

Additional processing adjustments could cause the apparent black region to expand or contract. Figure 9 shows the result of a Photoshop Curves adjustment applied to Figure 8; the adjustment expands the black region. An animation page^[L6] includes specific settings for the adjustment, a video of the full sequence from which Figures 7 and 8 were taken, and an animation that compares Figures 8 and 9.

Figure 9. A Curves adjustment expands the head shadow (Source: Figure 8, adjusted by the author)

Figures 10 and 11 are crops from the 4K scan. Figure 10 shows Frame 317, in which the shaded portion of Connally’s head appears light. Figure 11 shows the same head nine frames later, in Frame 326, with the shaded portion now appearing almost black after a change in head position and the resulting change in illumination across the head.

Figure 10. The shaded portion of Connally’s head in Frame 317 is lighter than it appears nine frames later in Figure 11 (Source: crop of the 4K scan)

Figure 11. Connally’s shaded head in Frame 326 appears darker (Source: crop of the 4K scan)

Figures 7, 8, 10, and 11 show that head shadows can vary substantially across frames as illumination and head position change. The corresponding line-profile graphs for each pair would likewise differ.

Open-Source Line-Profile Analysis Tool. Reproducing the studies cited as evidence of alteration requires line-profile software, yet most existing tools require either local installation and configuration or a paid subscription.

To give readers direct access, the author developed an open-source line-profile tool that runs entirely within a desktop browser.

In addition to plotting each RGB channel’s values along the line, the tool reports the mean and standard deviation of the absolute difference between each pair of color channels (|R - G|, |R - B|, and |G - B|). These statistics are explained in the Claim 1 section next, where they are first used. The tool can also analyze as many as three lines simultaneously.

Researchers can use the tool^[L7] to analyze any image they supply. Brief operating instructions appear when the tool is opened. All figures that use this tool include a link that opens it with the parameters applied to that figure.

Readers who want to independently compare the tool’s results against widely used tools such as ImageJ can run both tools against the same image and line coordinates. Readers may also examine the tool’s HTML source, which can be downloaded here.^[L8]

Claim 1: RGB-Coupling Difference

In a video shown by Mantik at the 2023 CAPA conference,^[L9] Whitehead discusses this specific claim. He states that a line through the shaded portion of JFK’s head shows loosely coupled red, green, and blue channel values before Frame 313 and more tightly coupled values afterward. He interprets the looser coupling in Figure 12, compared with the tighter coupling in Figure 15, as evidence that a black mask concealed the wound in the shaded area of the head. Paul Rutan, a postproduction expert, comments to Whitehead^[L10] (referring to the shaded area of JFK’s head in Frame 317), “It is just solid black.”

As noted earlier, the tool reports the absolute difference for each channel pair and reports the mean and standard deviation of that difference. Coupling has two relevant aspects, and each statistic measures one of them. The first is the average separation of the traces. The mean of |R - G|, |R - B|, or |G - B| measures the typical gap, so a smaller mean indicates that the channels run closer together. The second is the consistency of that separation along the line. The standard deviation of each difference measures how much the gap fluctuates, so a smaller standard deviation means the spacing is more consistent. Both are relevant here because Whitehead’s account addresses each. He reports that the three channels are distinct before the fatal shot and converge afterward. He attributes the erratic spacing he sees beforehand to Kodachrome grain, an account that implies the spacing steadies afterward.

A note on the use of statistics:

Whitehead’s claim that the channels converge predicts a decrease in their average separation. His additional attribution of the earlier irregular spacing to grain also makes the stability of that separation relevant. The mean difference therefore provides the more direct test of convergence, while the standard deviation provides a secondary measure of the reported change in spacing.

For compact descriptive comparison, the text occasionally reports a “mean total” or “standard deviation total,” defined here as the sum of the three pairwise means or the three pairwise standard deviations, respectively. These totals are descriptive summaries rather than standard statistical measures.

Frames 308 and 317 were selected for this article as representative of several before-and-after frames shown during the Whitehead discussion.

The RGB coupling along the line Whitehead drew across Frame 308, shown in Figure 12, is visibly looser than that shown in Figure 15.

Figure 12. Screen capture from a lecture video showing a line-profile graph for Frame 308 with noticeably loose RGB coupling (Source:[L11] 2023 CAPA)

Using a crop of Frame 308 from Figure 12, the author drew a new line slightly below the original line (Figure 13) and measured RGB coupling with the online tool described above. The RGB coupling shown in Figure 14 is noticeably tighter than that shown in Figure 12.

Figure 13. New line drawn slightly below the existing line (Source:[L12] analysis of a crop of the screen capture shown in Figure 12)

Figure 14. Graph and statistics of the line drawn in Figure 13 (Source: see Figure 13)

Figure 15 shows the RGB coupling along the line Whitehead drew across Frame 317. The coupling is tighter than that in Figure 12.

Figure 15. Screen capture from a lecture video showing that the graph of Frame 317 has tighter RGB coupling than the graph in Figure 12. Whitehead claims this difference indicates that a black mask was applied to several frames following the fatal shot, including Frame 317 (Source:[L4] 2023 CAPA)

Using a crop of Frame 317 from Figure 15, the author again drew a new line slightly below the original line (Figure 16). The RGB-coupling pattern in Figure 17 is broadly similar to that shown in Figure 15. The more relevant comparison, however, is between the author’s Frame 308 line and the author’s Frame 317 line. In that comparison, the two coupling measures move in opposite directions. The mean total is slightly higher in Figure 17 (6.40) than in Figure 14 (6.10), so the traces sit marginally farther apart rather than closer together. The standard deviation total is lower in Figure 17 (3.92) than in Figure 14 (4.99), so the spacing is somewhat steadier. Under that two-part formulation of Whitehead’s claim, both measures would be expected to decline; here, only one does. Neither the graphs nor the reported statistics show the pronounced two-part difference asserted in the claim. However, because both graphs derive from video screen captures, this comparison is indicative only.

The comparisons above do not isolate line placement from the limitations of the presentation images. Because Figures 14 and 17 are derived from video screen captures, differences in the measured coupling may reflect line placement, rendering and compression, or a combination of these factors. Line placement nevertheless remains relevant because changing the sampled path changes the pixels included in the measurement. Readers are encouraged to experiment with different line selections using the online tool.

A new line should not intersect the line already drawn on the image; crossing that overlaid line would severely skew the resulting graph and statistics.

Figure 16. New line drawn slightly below the existing line (Source:[L13] analysis of a crop of the screen capture shown in Figure 15)

Figure 17. Graph and statistics of the line drawn in Figure 16 (Source: see Figure 16)

These figures clarify one point while leaving another unresolved. In the available screen-capture images, a line drawn slightly away from Whitehead’s is associated with materially different coupling values, most clearly in Frame 308. Because these are presentation screen captures, the comparison cannot determine how much of that difference results from line placement rather than from rendering or compression.

Figures 18 and 19 therefore use a separately sourced scan. The late Roland “Rollie” Zavada, the Kodak film engineer who prepared the Zapruder film study for the Assassination Records Review Board, supplied the author with 4K scan files in September 2025, months before his death.^[L14] Graphs of these scans do not show the pronounced difference in RGB coupling reported for Frames 308 and 317. See the appendix for additional information about the 4K, 6K, and Museum scans, including the observation that the lower bit depth of the 4K and Museum scans could make dark regions more likely to appear flattened or masklike.

Notably, this potential bit-depth bias would, if anything, work in favor of Whitehead’s claim, since an 8-bit scan is more likely than a 10-bit scan to flatten dark values into an apparent mask. Yet the 4K scan shows no tightening of RGB coupling in Frame 317 relative to Frame 308.

The two 4K line-profile graphs in Figures 18 and 19 do not reproduce the reported difference in RGB coupling between Frames 308 and 317. As with Figures 14 and 17, the two coupling measures move independently, and here neither supports the claim. The mean total is essentially unchanged, declining from 23.00 in Figure 18 to 22.90 in Figure 19 (a decrease of approximately 0.4%). The traces therefore show essentially no change in average separation in this comparison. The standard deviation total rises from 11.08 to 13.40, so the spacing in Frame 317 fluctuates more rather than holding steady.

Whitehead’s two-part account predicts declines in both measures. Here, the mean is effectively unchanged, while the standard deviation changes in the direction opposite to that predicted by the account. Because the absolute values for Figures 14 and 17 derive from video captures, they should not be compared directly with those for Figures 18 and 19. The within-pair changes are nevertheless informative: the standard deviation falls in the screen-capture pair but rises in the scan-file pair. Differences in scan source, rendering, and line placement could account for that reversal. That sensitivity underscores the central methodological problem: without a controlled and reproducible procedure, the reported measurement does not provide a robust basis for determining what is present on the film.

Figure 18. Graph of cropped Frame 308 from the 4K scan (Source:[L15] analysis of 4K scan)

Figure 19. Graph of cropped Frame 317 from the 4K scan (Source:[L16] analysis of 4K scan)

Figures 20 and 21 are crops from the high-resolution Museum scan of a first-generation Zapruder film transparency (made directly from the original film). Because the author obtained only Frame 317 from the museum, the comparison between Frames 308 and 317 that is central to the claim cannot be performed with this scan. These figures demonstrate how substantially a single processing adjustment can alter the result. A single application of Photoshop’s Auto Levels command markedly increases both the measured separation between the RGB traces and the variability of that separation. The mean total rises from 21.30 to 56.30, and the standard deviation total rises from 9.76 to 24.41. Figure 21 differs from Figure 20 only in this one adjustment. This comparison does not establish which rendering is more faithful to the source; it demonstrates the sensitivity of the measured coupling to processing.

Figure 20. Cropped image and graph of Frame 317 from the Museum scan (Source:[L17] analysis of Museum scan)

Figure 21. Figure 20 after a single tonal adjustment (Source:[L18] Auto Levels adjustment of the image in Figure 20)

With a line-profile tool, the same line can produce substantially different graphs and statistics when processing parameters change or when a different scan is used. Conclusions drawn without the ability to reproduce all parameters may therefore be unreliable.

Film Grain

Whitehead comments on grain in his discussion of RGB coupling. He states^[L19] (author’s transcription):

So, as you can see there, they’re pretty much distinct red, green, and blue. I believe this is because we’re looking at the grains of Kodachrome as we go across, and they’re loosely separated here.

Claim 2: Light-Falloff Difference

The author of the slide central to this claim has not been identified. This article therefore refers to that person as the “unnamed analyst.”

Figure 22. Images and graphs produced by the unnamed analyst (Source:[L2] Midnight Writer News)

A. An Apparent Discrepancy between the Line Shown and the Graph

The publicly available slide contains an apparent discrepancy between where a transition occurs in the image and where the corresponding transition occurs in the graph.

In Figure 23, the green annotation line drawn on the slide appears to enter deep shadow at approximately 60% of its length, as measured along the line from left to right with Photoshop’s Ruler tool. A red line marks this transition. What appears to be the corresponding transition in the graph occurs at approximately 73% of the graph’s horizontal extent.

Given the slide’s low resolution, both measurements are necessarily approximate. Even allowing for several percentage points of measurement uncertainty, the two apparent transition locations remain noticeably separated. The uncertainty could be reduced only if the replication materials listed in the Independent Replication section were made available. The discrepancy is visible upon inspection; the measurement above merely quantifies it. Readers can repeat the measurement using the archived slide.

Figure 23. The red line added by the author shows where the green line enters deep shadow at approximately 60% of the line’s length (Source:[L2] Midnight Writer News; annotations by the author)

Figure 24. The graph places the beginning of deep shadow at approximately 73% (green-and-black striped line); the red line shows where Figure 23 places the same transition (Source:[L2] Midnight Writer News; annotations by the author)

B. Whitehead’s Falloff Claim about Connally

According to Whitehead’s interpretation of the unnamed analyst’s slide, Connally’s line-profile graph shows a more natural falloff than JFK’s graph does. Whitehead treats this difference as evidence that JFK’s wound was masked. See the graphs in Figure 22.

A line-profile graph shows how brightness changes along the sampled line, but those changes can result from more than lighting alone. They can also reflect the objects crossed by the line, the selected color channel, and image processing.

1. Additional Transitions

Connally’s graph may show greater variation partly because its sampled line crosses three features that JFK’s line does not: the limousine windowsill and two shaded areas of grass. These areas are visible in the 4K scan of Frame 317 in Figure 25 but are difficult to distinguish in the published high-contrast black-and-white slide in Figure 26.

Figure 25. Connally shown in a crop of Frame 317. The windowsill and two shaded areas of grass are visible (Source: 4K scan; annotated by the author)

Figure 26. Connally shown in a crop of Frame 317 from the unnamed analyst’s slide in Figure 22. The windowsill and two shaded areas of grass are heavily obscured by the overlaid line and the high-contrast rendering (Source:[L2] Midnight Writer News)

2. Scaling Issues

The slide in Figure 22 uses mismatched y-axis scales, complicating visual comparisons.

The slide lists the minimum and maximum channel-brightness values for each graph as 1,984 and 20,757 for JFK’s line and as 8,386 and 22,421 for Connally’s. These values yield y-axis ranges of 18,773 for JFK’s line and 14,035 for Connally’s. JFK’s plotted numerical range is approximately 34% larger than Connally’s. Because both graphs are plotted at the same height, Connally’s smaller range is stretched across the same vertical space, visually magnifying its apparent variation relative to JFK’s.

In the author’s comparisons, the JFK and Connally graphs use identical scales on both axes.

3. Color-Channel Issues

The slide in Figure 22 states that a single color plane, or color channel, was extracted, but it does not specify which channel was selected. Figure 27 shows variation across the color channels. This variation is particularly noticeable in the side-window assembly behind the front seat and in the heads of JFK and Connally.

Figure 27. Red, green, and blue channels (shown top to bottom) from a cropped version of the 4K scan of Frame 317, showing how brightness varies among the three channels for different objects (Source:[L20] 4K scan; channels extracted by the author)

Readers are encouraged to view an animation of the three channels.^[L20] The animation page also shows how line-profile graphs for JFK, Connally, and Greer change depending on the selected channel.

Because the slide states that a color plane was extracted, each graph appears to represent a single color channel rather than luminance.⁴ Luminance combines information from all three channels rather than isolating a single channel, thereby reducing the influence of channel-specific differences. Reliance on a single channel narrows the inferences the analysis can support, and the slide gives no reason for that choice. Here, the limitation is compounded because the channel is not identified.

4. Greer Graph Similarity

The author drew a line through Greer’s head using the same frame, scan, and processing used for the JFK and Connally lines, with Greer’s head in a comparable shadow. By visual inspection, the resulting graph more closely resembles JFK’s graph than Connally’s. No numerical similarity metric is offered here because the claim under review is itself framed as a qualitative difference in falloff shape; readers can verify the comparison directly using the linked animation and tool. Figure 28 illustrates the relative similarity of the graph shapes when luminance is used. As the channel animation^[L20] shows, this similarity to JFK’s graph (and dissimilarity to Connally’s) appears in each individual channel.

Figure 28. Luminance graphs from a cropped version of the 4K scan of Frame 317, showing that the Greer graph is more similar to the JFK graph than to the Connally graph. The order of the graphs is JFK, Connally, and Greer (Source: 4K scan; luminance computed by the author)

The four points above weaken Whitehead’s falloff claim. Among these observations, the Greer graph serves as a useful comparison: it resembles the JFK graph rather than the Connally graph, a result that complicates an explanation attributing JFK’s profile specifically to masking.

Methods

Links to author-supplied images that undergo line-profile analysis appear in an endnote.⁵ A printable list of all URLs is available here.^[L21] A small superscript containing the link number follows each hyperlink and is keyed to the URL cross-reference list.

Readers can determine the starting and ending coordinates for each line by clicking the figure’s source link and reading the coordinate values on the linked HTML page. Alternatively, readers may examine the URL to identify the coordinate parameters (e.g., bx=, by=, ex=, ey=, bx2=, by2=) for the line or lines drawn.

Images taken from YouTube videos or conference recordings are subject to video compression. Analyses using this material are generally confined to internal comparisons within a single image to reduce, although not eliminate, the effects of compression. When a comparison is made between two such captures, as with the statistics reported for Figures 14 and 17, the text characterizes the finding as indicative rather than conclusive.

Unless noted otherwise, figures identified as deriving from the 4K or Museum scans are based on the original scan files, so cross-scan comparisons use those files. Figures citing Midnight Writer News as the source rely on the images as published by that outlet.

The author used Adobe Photoshop 26.5.0 on a Mac to apply Auto Levels or Curves adjustments and to separate the color channels in Figure 27. Those channels also served as the basis for the channel animation. The author also used Photoshop’s Ruler tool to measure distances in various images.

The author used AI-assisted tools for proofreading and editorial review and then independently reviewed every incorporated suggestion.

Independent Replication

Conclusion

Author’s Note

If you have technical review comments on this article, please send them to craig@decodingmystery.com and keep them strictly focused on the material presented here. I will revise this article to correct any misstatements and will credit those who raise valid criticisms. I will keep all correspondents anonymous unless they tell me otherwise.

A printable list of all URLs is available here^[L28].

Appendix

4K, 6K, and Museum Scans

This appendix outlines pertinent similarities and differences among the 4K, 6K, and Museum scans.

During a 2024 visit by the author, Whitehead and Wilkinson confirmed the generational lineage of the 6K scan.

The 4K scan, supplied to the author by Zavada in September 2025, was made from a second-generation 35mm positive print.

On a printed presentation slide, Zavada added a handwritten note stating that one of NARA’s second-generation prints served as the source. In public presentations, Zavada stated that he arranged to have the prints scanned using Kodak Cineon-system equipment but did not identify the model. If the Cineon scan followed the standard 10-bit workflow, the files the author received from Zavada would have been reduced to 8 bits at some later point in the chain. The bit depth of the original scan could not be confirmed.

The Museum scan has a different generational history from the 6K and 4K scans. Specifically, Joseph G. Barabe of McCrone Associates made 4-by-5-inch transparencies from the camera-original film at NARA in 1997 as part of the Zapruder Family / MPI Home Video preservation project; they were subsequently scanned for the Image of an Assassination DVD release.^[L22] The DVD states that the scans were made at 1,500 dpi. Given that the image area was approximately 4 inches wide, the original pre-DVD scan spanned roughly 6,000 pixels. Standard-definition DVD-Video, however, uses only 720 horizontal pixels; because select scenes undergo fourfold magnification to focus on the limousine, those frames retain the equivalent of roughly 2,880 pixels of detail across the full-frame width.

The museum later scanned these transparencies. The author purchased a high-resolution scan of Frame 317 from the museum on October 2, 2025. According to the file’s metadata, this frame was scanned in 2010 and is 23,500 pixels wide.

Both the 4K files supplied to the author and the Museum scan are 8 bits per channel. According to Whitehead, the 6K scan is 10-bit log-encoded. Compared with an 8-bit display-referred file, a 10-bit log-encoded file can preserve a wider range of tonal information, particularly in deep shadows and highlights, although the practical advantage depends on the scan and subsequent processing.

The Museum scan has by far the largest pixel dimensions. Given the finite resolving power of the camera-original Double 8 Kodachrome II stock (approximately 100 line pairs per millimeter under ideal conditions and with optimum equipment), these differences in pixel dimensions have little bearing on recoverable scene detail because all three scans sample the film beyond its resolving power.

The 6K scan’s higher bit depth gives it an advantage in tonal sampling. The lower bit depth of the 4K and Museum files makes adjacent dark values more susceptible to quantization, which can cause them to merge into identical code values. This loss of tonal differentiation can make dark regions appear flatter and, depending on the color-processing pipeline, can make channel traces appear more tightly grouped. All else being equal, the 8-bit files may therefore be more likely than the 10-bit file to produce a masklike appearance in deep shadow.

Ultimately, the disparate shadow densities across the renderings in Figures 1 through 6 reveal a simple reality: processing choices materially shape what the viewer sees.

Zapruder Film © 1967 (Renewed 1995) The Sixth Floor Museum at Dealey Plaza.

Fair Use Notice. This article reproduces limited portions of copyrighted works for purposes of criticism, comment, scholarship, and research under 17 U.S.C. § 107. The excerpts are used solely to identify and evaluate the specific claims under discussion and are published without charge. All rights remain with their respective owners.