Later I had access to Canon CanoScan 2700F and Polaroid SprintScan 4000 scanners and used them to scan the same test image for comparison. Canon costs approximately 1/4 of the price of Agfa but it offers higher maximum resolution while Polaroid is more expensive has maximum resolution of 4000 PPI.  Juha Helminen was kind enough to scan my test slide with his Hewlett-Packard PhotoSmart S10.

Figure 1. Full 35mm frame used in this test, photographed using AF Nikkor 85mm f/1.8 lens stopped down to around f/5.6 and Fuji Provia 100F (RDP-III) film. The image was mounted in 3mm Gepe mount with antinewton glasses when it was scanned. Colourful rectangles are highlighting areas used below.

Figure 2. Comparison of different resolutions used in this test seen 1:1. Only Autolevels performed in Photoshop. If you look closely you can see how jaggedness of number 1 in the price tag reveals the asymmetric resolution of 2000 PPI Agfa scan. Canon looks very smooth but rather soft, HP shows much more detail.

 

Figure 3. Same data as above but lower resolutions resampled to 2720 PPI in Photoshop. It's obvious that despite Canon's higher resolution HP has extracted much more information out of the image. Interestingly Agfa looks better at 1000 PPI than 2000 PPI!

Figure 4. Three different parts of the image with higher resolution scans resampled down to 1000 PPI. Unsharp mask filter applied to all images in Photoshop. 1000 PPI Agfa scan clearly has the edge here, resampling seems to cause some softening which is not restored by unsharp masking, only in the middle row 2000 PPI Agfa scan looks better than 1000 PPI. HP is missing because I already forgot parameters used for unsharp masking.

I wanted to know how close to the maximum resolution of this film scanners could get.  I didn't have access to a microscope with photographic possibility, so I ended up projecting the slide with Leitz Colorplan 90mm f/2.5 lens and photographed the projected image with Olympus C-2000Z digital camera. As I couldn't have the camera on-axis with the slide projector I stopped Olympus down to f/8.0 to ensure enough depth of field resulting in exposure of 1/2 seconds, so I had to use tripod and remote control release for maximum quality. 

Figure 5. From this comparison alone, it's obvious that the slide has much more resolution in it than what the Canon scanner can get out of it. HP on the other hand is getting pretty close, only the finest details are missing. Lowest image is strecthed horizontally to overcome compression caused by off-axis placement of the camera.

 

Figure 6. In this comparison the effective resolution of the photograph is around 4080 PPI. Sharpened Polaroid scan looks slightly softer by comparison, it's obvious that at highest resolution it's not producing as sharp results as one could hope for. HP image is surprisingly good despite the lower resolution. Enhanced contrast in the photograph makes comparison more difficult, but film grain is clearly not real problem yet. From this comparison alone I'd argue that at least 4000 PPI resolution is needed to get all the useful information out of 35mm slide. In terms of digitals cameras this is roughly 20 megapixels.

Figure 7. Comparison of noise found in dark colours of the image. Canon allowed only 24-bit output file. Agfa and HP image was scanned with higher colour resolution and converted down to 24-bit in Photoshop before applying "Equalize" making it easier to see the noise.

My conclusion: I will happily continue using Agfa DuoScan at the resolution of 1000 PPI to scan my slides for web use as it proves to have good signal to noise ratio both resolution wise and colour wise. HP PhotoSmart S20 proves to have superior resolution compared to Agfa, also outperforming Canon older model with higher maximum resolution by wide margin. Polaroid produces very good results with 2000 PPI, but at 4000 PPI it's not as good as I hoped for.

Before drawing any further conclusions bear in mind that this test was not executed with the precision of a real scientific study. Any comments regarding to this scanner test can be sent to Petri.Kekkonen@oulu.fi