Film Flatness Test: 3 Film Holders in the First Quantitative Comparison

This article documents an attempt to quantitatively compare film flatness in three DSLR scanning film holders. The method is based on depth-from-focus analysis using a macro focusing rail. Measurement accuracy is limited; the results show trends and relative differences, not absolute values.

Why film flatness matters

Film flatness is the most common complaint in the DSLR scanning community. The film bows in the holder, and the corners lose sharpness. The physics behind this is well understood: With typical scanning settings, the depth of field is only fractions of a millimeter. A bow of 0.3 mm is enough to produce visible loss of sharpness in the corners.

Many opinions, no measurements

Forums and Reddit contain numerous claims about film flatness: “The corners are soft,” “Holder X is better than Y.” What you don’t find is measurement data. In our research, no one had quantitatively determined and published a film’s deviation in a holder. All comparisons are based on subjective assessments of scan sharpness.

This article is an attempt to change that.

The test strip

The test strip used was Kodak Gold 200, shot with an Olympus OM-2n and Zuiko 50mm f/1.8 at f/5.6. A relatively dense negative with evenly distributed image information.

The same strip of film and the same frame were used for all measurements. The choice of subject matters: The depth-from-focus method needs texture in the image (film grain, edges) to determine the point of focus. An evenly bright or dark subject (sky, wall, snow) provides no usable signal. Branches and leaves work well: dense, high-contrast, and evenly distributed across the entire frame.

Method A: The reflection check (qualitative)

The film surface reflects light. When the mounted film is held under a ceiling lamp, irregularities become visible as distorted reflection lines. This method is simple and reveals fine details, but they cannot be quantified.

The same strip of film in four holders, photographed under the ceiling lamp each time:

Reference: The same film held by hand, without a holder

What you can see with the naked eye: None of the holders keeps the film perfectly flat; the reflection lines are distorted in all three. But all three visibly perform better than no holder at all. At best, you can guess which performs best from the reflection images. The curvature cannot be quantified; the reflection does not reveal whether the deviation is 50 or 500 micrometers. That is why the second method is needed.

How the three holders guide the film

The three holders tested use different design principles to keep the film in position. This affects where and how strongly the film can warp.

Negative Supply: The base has a panorama-format gate, and the film is guided only along its two long sides (top and bottom). There is no guidance along its length. Masks are available that slide in from below (for Half Frame, 35mm, or panorama), but even with a mask, the film remains unconstrained lengthwise. This means that the film has the most room to sag in the middle of the gate.

Valoi: There is a separate, one-piece molded holder for each film format. The film is guided both across its width and along its length. The design of the longitudinal guide cannot be seen from the outside because it is integrated into the housing.

Ausgeknipst: A combination of both approaches. Like Negative Supply, the base has a panorama-format gate and guides the film only at the sides. Interchangeable tops are placed on it to guide and press the film from above along its length as well. Without a top, the holder behaves like the Negative Supply approach (side guidance only). With a top, longitudinal guidance is added.

What this means for the film edges: All three systems guide the film exactly along the area that would have to be exposed to scan the perforations and edge markings. A holder that reveals the edge can no longer grip the film there, and the control measurement further down shows where this ends: without the top plate, the same strip with 381 µm RMS deviates from the plane 2.3 times more. Edge scans and flatness are mutually exclusive in this design. Anyone who wants both cannot avoid using two Anti-Newton glasses, and pays for it in speed because each frame must be inserted individually.

This difference in the guide is relevant to the measurement: Since the gate width differs for each holder, the analysis was limited to the inner 80% of the film frame. The edge areas contain little image information and are cropped out during scanning anyway, so they are not included in the comparison.

Method B: Depth-from-focus measurement (quantitative)

The camera moves through the film plane on a macro rail in defined increments. An image is taken at each position. Each region of the image reaches its maximum sharpness in a different frame, depending on how far it is from the lens. The height of the film surface at every point can be calculated from the position of the sharpness maximum.

Setup

Measurement setup: camera on a macro rail, film holder on the light source. The Post-it notes under the lightbox serve as shims, compensating for minimal height differences at the corners so that the film plane is parallel to the sensor.

Close-up: macro rail with dial indicator, 0.1 mm increments

The details:

• Camera: Sony ZV-E10
• Lens: Carl Zeiss Jena Tessar 50mm f/2.8 on a macro bellows
• Aperture: f/2.8 (wide open, for maximum sensitivity to focus shifts)
• Macro rail: 0.1 mm increments, 21 shots per pass
• Alignment: Mirror method

Mirror alignment: The lens reflection must be exactly centered so that the optical axis is perpendicular to the film plane

Analysis

A Python script performs the analysis. It divides each image into a grid and determines which frame has the highest sharpness for each cell. This makes it possible to calculate the height of the film surface at every point. Two corrections follow: First, the global tilt is removed (the sensor is never perfectly parallel to the film). Second, the lens field curvature is removed so that only the film's actual topography remains.

Three passes per holder, shuffled and averaged.

Technical details of the data processing

For the analysis, the script divides each image into a grid of 20 x 30 cells. For each cell, the Laplacian variance is calculated across all 21 frames, a measure of how much high-frequency contrast (film grain, edges) the respective cell contains. The cell is sharpest when the focus plane lies exactly on the film surface.

The resulting sharpness curve has one peak per cell. Its position is determined using three-point parabolic interpolation to achieve finer Z resolution than the step size (0.1 mm). A best-fit plane is then fitted to the entire Z map and subtracted (tilt correction). When comparing multiple holders, the mean of all Z maps is subtracted as system bias (common-mode rejection), removing the lens field curvature that is identical across all measurements.

The analysis covers the inner 80% of the film frame. The edge areas are discarded because they contain little to no image information and are cropped off during scanning anyway.

The scripts are written in Python (numpy, opencv, matplotlib). Anyone who would like to review the code or recalculate the raw data can get in touch; if there is justified criticism of the method, we will adjust the analysis and publish the corrected results.

Worst-case reference: film guided only at the edges, without a cover plate, with maximum sag

Results

Three runs per holder, averaged. Measured across the inner 80% of the film frame (edge areas excluded because the holders guide the film at different widths; see above). Lens field curvature removed using common-mode rejection.

PV = Peak-to-Valley: the greatest deviation between the highest and lowest point.
RMS = Root Mean Square: the average deviation, significantly more robust than PV because a single outlier does not distort the result.

Heatmaps

The heatmaps show the topography of the film plane. Red means the film curves toward the lens. Blue means it curves away. White is the ideal plane. The scale shows micrometers.

Three-holder comparison: Red = film closer to the lens, blue = farther away

Validation: The sprocket holder (left) clearly shows more deviation than the holders with top plates

And the individual heatmaps, each averaged over three runs:

Ausgeknipst, averaged over 3 runs

Valoi, averaged over 3 runs

Negative Supply, averaged over 3 runs

What the data says

Validation: Does the method work?

The sprocket control (film without a top plate) shows RMS values 2.3 times higher than the best holder. This is the most important data point of the entire measurement: It confirms that the method resolves real differences in film flatness and that the results are not lost in measurement noise.

Ausgeknipst (RMS 163 µm)

The lowest RMS value in the test. The heatmap shows a relatively even distribution without dominant hotspots. The interchangeable top presses the film both along the sides and along its length, which is reflected in consistent guidance. The run-to-run variation (how much the results differ when the same film is loaded repeatedly) was 176 µm, comparable to Negative Supply.

Valoi (RMS 175 µm)

Just behind Ausgeknipst. The heatmap shows slight wave patterns that could be due to the S-curve guidance of Valoi's channel design; the film is guided through a curved channel when inserted. The run-to-run variation was slightly higher than with the other two holders at 210 µm. The data cannot determine whether this is due to the design or the way the film is threaded.

Negative Supply (RMS 202 µm)

The highest RMS value of the three holders. The heatmap shows more contrast than the other two, with areas of greater curvature appearing more pronounced. Negative Supply guides the film only along the long sides, not along its length. The missing longitudinal guidance could explain why the deviation is slightly higher in the middle of the gate. At the same time, the run-to-run variation was the lowest in the test at 175 µm, meaning the film sits consistently in the same position each time it is loaded.

The comparison

The ratio between the best and worst holder is 1.2x (163 vs. 202 µm RMS). In absolute terms: a 39-micrometer difference. That is less than the thickness of a human hair.

For context: At f/8, the aperture at which most people scan, the depth of field at the negative is approximately 500 micrometers. All three holders keep the film within this tolerance. The 39 µm difference will not be visible in a finished scan at this aperture.

At wider apertures (f/4 or f/2.8, as used in high-end scanning setups), the depth of field shrinks to below 200 µm. In this range, the measured differences could become relevant, but even then, the effect is difficult to separate from other sources of error (sensor alignment, lens field curvature, curl of the specific film strip).

Conclusion

All three holders keep the film measurably flatter than an unsupported strip of film. The differences between them are small; the ratio between the best and worst result is 1.2x.

The products differ in many other respects (material, workflow, compatibility, price). Film flatness is one factor among many. For this one factor, the three tested holders are close to one another.

Setup note

Measurements were taken at f/2.8 (wide open). This is not usual for scanning; at wide-open apertures, lens performance decreases, especially in the corners. The reason for using the wide-open aperture is that the depth of field must be small enough for measurable differences to occur in the film plane at all. At f/5.6 or f/8, the DoF would be too large to resolve film curl. A 100mm macro lens at 1:1 magnification and slightly stopped down would have been the better measurement instrument, but it was not available.

To the community: Help us measure better

This experiment was a first attempt with limited equipment. The method has weaknesses, which are documented above. If anyone in the community knows of a more precise, affordable method (laser interferometry, moiré topography, or something else), we would appreciate the suggestion. The tests will be repeated, and the raw data published.

The goal is not a marketing comparison. The goal is to improve our own design based on measurement data.