The original article, Camera Scan vs Film Scanner, showed camera scanning using a Canon EOS 7D, which has an APS-C size 18 Mpx sensor. The lens was EF-S 60 mm f/2.8 USM Macro. It demonstrated how camera scanning compared to a dedicated film scanner. Camera scanning was superior, but there were still some limitations in deep shadow performance, and perhaps also resolution – fuzzy grains and marginal sharpness in the corners. A follow-up article, What is the optimal aperture for camera scanning?, discussed theoretical considerations on what the optimal aperture should be as a trade-off between depth-of-field and diffraction.
Now I have a new camera with both higher resolution and a higher dynamic range: a Canon EOS 80D, still an APS-C size sensor, but with very significant improvements. I also have a new high-end macro lens: Sigma 70mm f/2.8 DG Macro Art, with sensational performance.
This article is about the quality improvements the new equipment brings, and it also contains recommendations based on very deep studies into how to achieve the best results.
[Stockholm, March 2019]
Comments on this article are welcome at the posting on photo.net.
The setup is economical enough to be a viable solution for ordinary quality-minded people. The camera is a Canon EOS 80D, which has an APS-C size sensor with 24 Mpx. The lens is a high-end macro lens from Sigma's Art line, designed for full frame: Sigma 70mm f/2.8 DG Macro Art.
Since the sensor in EOS 80D has 6000 pixels horizontally, that equates to 4233 ppi (pixels per inch) when camera scanning a 24 x 36 mm slide. That matches very well what is possible to extract from such a slide, including grains.
In the article Camera Scan vs Film Scanner, the camera was a Canon EOS 7D, which was introduced in 2009, which is almost 10 years ago. The EOS 80D was introduced in 2016. Although 80D is regarded as a step below the pro-level "single digit" cameras, its sensor is a very substantial improvement over 7D. A Mark II version of 7D was released in 2014, but the sensor in 80D is significantly better, and it has a fully articulating touchscreen.
The most obvious difference is of course its higher resolution, 24 Mpx vs 18 Mpx in 7D (20 Mpx in 7D MkII), but much more important is that its dynamic range at low ISO is much higher than that of the 7D sensor. The higher dynamic range manifests in an ability to raise shadow details without experiencing banding or other artefacts from electronic noise. This enables us to capture the entire range of tones in high-contrast slides, from the very deepest shadows to the brightest highlights. Another significant feature is the dual-pixel sensor, which makes Live View enormously much more useful since its autofocus is so much better than 7D in Live View. There are also several other aspects in which the 80D is better than the oldish 7D, but those things are not highly relevant for the use as a camera scanner.
Sigma's top-of-the-line lenses are labeled Art. This lens is a new macro lens as of 2018, with sensational performance. It is full frame (FF) so it works very well with both FF and APS-C cameras.
There is an advantage in using an FF lens despite that the camera sensor is APS-C: we hit the "sweet spot" with an exceptional performance from corner to corner. With camera scanning, this is significant – we demand that the grains look the same throughout the entire slide from edge-to-edge. And we cannot tolerate any field curvature at all since the depth of field is only a small fraction of a mm. We also demand zero distortion (horizontal and vertical lines must be perfectly straight lines). My previous lens, Canon EF-S 60 mm f/2.8 USM Macro was very good, but the new Sigma is better: The difference is that the Sigma has no detectable field curvature, and the sharpness at the edges are virtually the same as in the center, it has virtually zero distortion and zero chromatic abberation. At f/5.6, the vignetting is extremely small. In the corners of APS-C, it reaches maximum resolution already at f/4.
The Sigma Art macro has another huge advantage: Its focus ring is "focus-by-wire", which means that the focus ring controls a motor that drives focus; the focus ring has a very large angle of rotation (about 2⅓ turns with normal rotation speed, and if you rotate it slowly it goes into fine-tune mode); this allows you to focus very accurately. With EF-S 60 mm f/2.8 USM Macro, a tiny touch of the focus ring would through focus way off – virtually impossible to set the focus accurately on the grains. With the Sigma 70mm f/2.8 DG Macro Art, I can generously turn the focus ring smoothly to set the focus on the grains!
With the EF-S 60 mm f/2.8, all focusing was with AF. There is a problem though to let AF set the focus in a camera scan: First, AF itself is not 100% consistent, there will always be small variations, but our margins are extremely small, so even those small variations often cause unsharpness in some part of a bulging slide. Second, since a slide never has any crystal-clear high-contrast edges to focus on, I suspect that is one of the reasons that AF can vary a bit too much – the slide itself is fuzzy, and even the grains are a bit fuzzy. Third, I suspect the AF can be mislead by dust particles that extend a tiny bit above the surface, and if these dust particles are on the film-base side (not the emulsion side) it makes things worse. These are probably the reasons I had to stop down pretty much to get edges of the slide reasonably sharp.
I have come to the conclusion that the only safe way to focus is to focus manually on the grains in Live View! The Sigma Lens is excellent for that! In the camera, I set the Picture Style (a Canon specific thing mostly intended for jpg) to maximum sharpness with no threshold. That makes the grains more pronounced in Live View with almost halos around the grains. The final image is in RAW format so the picture style is ignored.
I focus on a part of the slide somewhere between the center and the edge, with smooth color where the grains are relatively clearly visible. This will make the area of optimal sharpness more like a ring rather than a point in the center of the bulging slide. DoF will be better exploited that way. f/5.6 is the optimal aperture, but it may happen that the grains close to the edges are softened due to a limited DoF. But since the resulting image usually requires a little bit of cropping, this might perhaps be acceptable. f/6.3 improves DoF a tiny bit but also increases diffraction a tiny bit (which, however, can be compensated by increasing sharpening a bit).
I use the same old German Novoflex slide duplicating attachment, originally designed for a bellows system, that I modified and adapted for this use.
As illumination, I use the same standard Canon flash that communicates with the camera to provide just the right amount of flash power for the current slide – i.e it adapts power depending on bright/dark slides. This flash targets a mirror in front of the setup, angled so that the flash bounces back right onto the slide duplication film/slide holder with its opal diffusing glass. I use a regular laser pointer to find the right distance and angle to achieve the right angle of the mirror.
The whole setup is mounted on a rail (which makes detaching the camera a lot easier than the simpler setup I had originally with 7D and EF-S 60/2.8, and it's also significantly sturdier). I had to adapt the whole slide holder equipment for this, both in height and the way it attaches the rail.
It is extremely important to line up everything so that the slide lines up exactly perpendicular to the lens focal plane! That is difficult! Since the depth-of-field is only about ±0.2 mm (at roughly f/5.6), the four corners of the slide may not differ more than a fraction of that value, since we need the DoF for the bulging of a slide and cannot afford any misalignment of the slide. Therefore, we should target a precision of less than ±0.1 mm at the slide corners! That is difficult!
I measured with a good vernier caliper (which has 2/100 mm resolution) from each corner of the slide frame to the lens rim. (The slide frame is 5 x 5 cm) I got it down to a difference of about 0.14 mm after considerable work, which should translate to ±0.044 mm at the corners of a 24 x 36 mm slide. I discovered that the slide holder (which can slide up and down for vertical adjustments) is itself not fully parallell to the rest of the Novoflex equipment – I had to use aluminum foil as a shim under the plate that attaches it to the bigger black metal plate in order to resurrect parallelism. I can't guarantee the above claimed accuracy, but it should in any case be less than ±0.1 mm.
The slides are either mounted in glassless Gepe frames, or they were delivered in plastic or paper frames from the lab. The films are mostly Kodachrome II, K25, K64, and Fujichrome Sensia. Before each scan, the slide is meticulously dedusted under strong sideways light and a good loupe. The resulting RAW files, about 30 MB each, are imported into Lightroom.
I made extensive tests scanning the slides either straight (the way you look at the slides, with the film base facing the lens) or reversed (mirrored, with the emulsion facing the lens). This was primarily done for the purpose of finding out if the lens has any field curvature, so I used full aperture. Since slides usually bulge so that the emulsion side is concave, one might expect that reversed slides would be best if the lens has a field curvature. However, I could not detect that with this lens (but I could do that with Canon EF-S 60 mm f/2.8 Macro). With reversed shots, I simply flip them horizontally in Lightroom.
With AF, I had somewhat greater success with reversed slides – more consistent sharpness throughout the slide. This may be difficult to explain, but I suspect that it could be due to any or all of the following: (1) dust on the film base might easier mislead AF if the film base faces the lens; (2) the film base itself might somehow make it more difficult to AF as compared with the naked emulsion; (3) some people claim that the film base diffuses light somewhat, which would only be a problem if the film base is between the emulsion and the lens; (4) if any light is reflected back from the lens to the slide, the film base would reflect that back to the lens again.
Considering all of the above, I have come to the conclusion that it is safest to camera scan the slides reversed (emulsion towards the lens), even if I focus manually on the grains. This way I am guaranteed that there is nothing in the light path from the emulsion to the sensor, and if the lens has any field curvature at all it should be better if the slide bulges out from the lens. I can not see any disadvantages with this method, other than I have to flip all images horizonally in Lightroom, but that is very easy.
To my great surprise, calibration profiles have wildly different effects when applied to files from the 80D compared to files from the 7D: For instance, the difference between Adobe Standard and Camera Faithful with files from the 80D is opposite to the difference with files from the 7D! I have had to completely remake all my settings to create new defaults for 80D!
I have had to completely abandon the special slide presets I had for 7D that set White Balance, Tone Curve, Shadows and Color adjustments to make the images on the monitor match as closely as possible the appearance of the slides on my light table that uses 5-band 5000 K fluorescent tubes.
I made new separate presets for KII/K25, K64, and Sensia. They are based on Adobe Standard. (I worked for a long time with Camera Faithful, but I had to abandon that profile since it has severe problems with certain dark red transitions.)
With 7D, I had the same preset for Sensia and KII, but that does not work well with 80D. I can't explain that, and it is surprising that I get so different results for different films. KII and K25 are sufficiently similar though. Ideally, one would like to think that basic good defaults for general photography would also be good for camera scanning, but that is obviously not anywhere near the truth.
The presets have adaptations in the panels Basic, Tone Curve (both point curves and parametric) and calibration. A lot of time-consuming work! The end result, however, is an extremely close match between the images on the monitor and the appearance of the slides on the light table!
All images below are clickable and open in popup windows with (if needed) scrollbars. Inside each window there is text at the bottom saying what it is and its resolution. This way, you can compare different images by switching windows or placing them side-by-side. Some are downsampled, some are 1:1 images, i.e 100% pixel zoom. The images in the popup windows do not scale to your monitor – this means you can rest assured that the image size of these images are exactly as I have set them. All of them have been converted to the sRGB color space on export for this web page.
These two images are camera scans from 80D and 7D respectively. The slide is Sensia 100, shot with a Canon EOS 5 with an EF 20-35 mm f/2.8L. That slide is chosen as a critical test because it is an unusually difficult one: It has specular highlights, normal highlights of well known gradations, well known texture in water and buildings, and most importantly: it has details way down in very deep shadows. Click on them to compare colors. (The photo was taken in 2003 in Istanbul from the Galata Tower.)
The top image is the EOS 80D camera scan obtained as described above. It matches the slide extremely well in all aspects as viewed on the light table! Even the perception of the grains is preserved. The bottom image is the EOS 7D camera scan obtained as described in Camera Scan vs Film Scanner (slightly color balanced here for this web page).
The red rectangle imposed on the top image shows a detail of the picture that we will take a closer look at in 1:1, i.e 100% pixel zoom:
When you click on these images, they will open in popup windows with (if needed) scrollbars. Inside the window there is text at the bottom saying what it is. They are not downsampled, i.e they are 1:1 images – one camera pixel being one image pixel in these jpgs. The images in the popup windows do not scale to your monitor – rest assured that the image sizes of these images really are 1:1 pixelwise, and they are all converted to the sRGB color space on export for this web page. (For those of you who use a Mac retina monitor, or similar type, the browser will render 1 image pixel onto 4 screen pixels (or dots if you like) when the zoom is set to "Actual "Size". The physical size of the image will therefore be the same on an old style "ordinary" monitor as on a high-resolution retina monitor. If you want to see the image so that 1 image pixel is mapped to 1 screen (retina) pixel (or dot), just as Lightroom does at 1:1, you will have to zoom out.)
The top one is the 80D camera scan, i.e a 1:1 crop of the image shown in the previous section. It matches the slide extremely well as viewed on the light table! It has had both sharpening and noise reduction applied. The subjective perception of the grains is well preserved.
The next one is the 7D camera scan, i.e a 1:1 crop of the image shown in the previous section. It does have a slight tendency to banding in the very darkest portions, and there is some dark "haze".
Start by cranking up brightness on your monitor. Then click on the first two images to compare the two camera scans by switching windows back and forth. If needed, scroll them up so you can see the descriptive text. Look at the lower part of the building in the middle, which has vertical stripes and very dark windows. The 80D camera scan brings out the deep black details significantly better in these deep shadows! Granted, there are some color differences, and it is hard to say which ones match the slide best.
As for sharpness, there is a significant difference in the sharpness of the grains, and the 80D scan might even look a bit over-sharpened, but it actually has less sharpening applied than the 7D scan! Try the zoom-out function in your browser to study the images in less than 1:1 so as to judge the general perception of grains and deep shadow performance in a more realistic viewing condition.
[Update: Since this article was published, I have reduced the sharpening (in the 80D scan), since I think the grains really were over-sharpened, artificially pronounced, which is especially disturbing in portraits. The reduced sharpening had insignificant effect on the actual image details, but it made the grains more natural the way they are perceived in the physical slide.]
As for image detail, there may not be much of a difference between the two camera scans, but the 80D appears to be a bit more "convincing", or a bit more pronounced, in rendering the details. If the slide had been Velvia shot with a prime lens, the difference may perhaps have been a tad larger.
If you compare the 80D camera scan with the one scanned in the film scanner ArtixScan with SilverFast 6.5, as described in Camera Scan vs Film Scanner, it is obvious that the 80D camera scan is way superior! The difference is dramatic, the camera scan easily beats the multisampled ArtixScan which becomes murky in the shadows.
In summary: Although the 7D camera scan was a bit better than the film scanner I had, its sensor from 2009 is not surprisingly beaten by the 80D, and with the new Sigma macro Art lens, the combination accomplishes an end result that has captured everything the slide contains!
As before, all images are clickable and open in popup windows. At the bottom of each window, text says what it is and its resolution.
These two images are an 80D camera scan and a 7D camera scan respectively of a Sensia 100 slide. (The photo is the sunset at Selfjorden on the northwest side of Moskenesøya, Lofoten, Norway, in July 2007. The lens is an EF 200/2.8L USM)
The top one is the 80D camera scan. It matches the slide extremely well as viewed on the light table! The bottom one is the 7D camera scan.
Both do a very good job, but there are some color differences; I regard the 80D version as a better match to the slide on the light table. The 80D also delivers significantly better shadow performance without any "shadow haze".
A camera scan using a Canon EOS 80D (24 Mpx) with a really good macro lens is capable of extracting everything from a high contrast slide with a color rendition that is extremely close to the subjective perception of the slide as viewed on a light table that uses 5-band 5000 K fluorescent tubes.
However, it is not so easy to get good results by simply "taking a picture" of the slide. First of all, the depth of field is so small that it can be very tricky to get the slide absolutely parallel to the focus plane of the lens. And to achieve the same subjective appearance of the image on the monitor as compared with the slide on the light table, quite a lot of tweaking to contrast and color balance has to be done. And the different properties of films makes it necessary to develop separate presets for them.