Pentax67+Velvia vs Sony Alpha 900

Tuesday, 18 August 2009 21:45 administrator
User Rating: / 0



Readers Comments

Another way to see it

Update 2009_11_08

Update 2009_10_30

Evaluating with Imatest (updated)

Comparing Unscaled Images

Comparing Scaled Images


Planned tests


To begin with

I shot a comparison between my new digital camera a Sony Alpha 900 and my old Pentax 67. The comparison was shot on Fujichrome Velvia 100 and scanned on a Minolta Simage Scan Multi Pro slide scanner. This article is work in progress, more images and info will be added when available.

One of the reasons I made this test is that I find my images scanned at 3200 PPI quite ugly, noisy, lots of grain, not very sharp and more often than not significant chromatic aberration. This is not visible with a 15X "Peak" loupe nor in my Götschman projector, expect perhaps when looking at a projected image from very close. For that reason I wanted to compare with digital. I only made a few large (70x100cm)  prints from 67 slides, but those impress professionals who have been working at some of the best professional labs in Sweden, so I suggest that the image quality we discuss is high, but we are looking into tiny details. These prints were based on scans from slide film using the same methods as employed for the tests and printed on photographic paper on Durst Lambda without operator intervention. (The Swedish lab Crimson has a service for professional photographers offering prints without operator intervention, half the price, fast turnaround, no one but yourself to blame!)


I'd also point out that this is an evaluation of the equipment I use. Others using other equipment may achieve other results. Also, low speed black and white film has much higher resolution than the Velvia used here.

I have seen few comparison between my scanner and more expensive stuff, this is probably the best one:




Update 2009-11-08

I paid some consideration to the tought of using a macro lens with bellows and photograph a small part of the transparency instead of scanning it. Unfortunately this was no great success. The aperture control does not work with my Novoflex bellows any longer, so I could not control aperture using minimum aperture was ruled out because of diffraction. I may try to use the bellows with an enlarger lens instead. Using the Macro 100/2.8 at 1:1 gave a result which was very similar to the scanner on the red flower image. The red flowers had the same behaviour as with the scanner. Not sharp and little tonal separation of nuances.


Another investigation I made was to compare my $-bill to the $-bill from the great MFDB shootout of 2006. The reason I was looking into this was that there were no 24 MB DSLRs back in 2006. In the MFDB shootout there was Fujichrome Velvia in 645 and 4x5" included. The Velvia slides were drum scanned by Charlie Cramer. No qustion that Charlie Cramers drum scan from the &45 Velvia was better than my scan from Velvia 67, if it was sharper may be different. Anyway, it's obvious that better scans than I have achievable.


Comparing to a test by Luminous Landscape

Some very good testing has been done by the gentlemen Bill Atkinson, Charlie Cramer and Michael Reichman a couple of years ago. The test was published on Luminous landscape and also made available on a DVD. The image below shows their drum scanned Velvia image shot on a Hasselblad H1 on the left, my scan on the right. The LL-image takes less sharpening. Something that can be noted that the LL image does contain "shadow detail" that is lost in my Velvia scans but visible on the DSLR image.


The latest tests (2009-10-30)

Klick here to download the original images...

A series of tests were done to identify possible issues with focusing. It was established that some small focusing error probably exists, on the order of 0.5 cm in 1 meter with 80 cm lens on the Pentax. The Sony is used with "microfocus adjustment", so it's focus is essentially correct. Scanning was done with my Minolta Diamge Scan Multi Pro at 4800 PPI. This scanner is capable of scanning MF at 3200 PPI and 135 at 4800 PPI. 4800 PPI can be used with MF but covers only the central part.


Two comparisons were made, one at about 12000 Pixels width, which is the native resolution of the scanner used. The scanned picture here is not resized while the DSLR image is rescaled using bicubic interpolation. The comparisons were made with either image unsharpened (top row) and with the scanned image sharpened at nominal resolution and the DSLR image capture sharpened using the "landscape preset". The scanned image was sharpened as follows (Amount 163, Radius 2.3, Treshold 10). The unsharpened images are shown in the top row and the sharpened ones in the bottom row. Velvia left and DSLR to the right.


The other comparison was based on 8000 pixels width. The Velvia images were downscaled using bicubic sharper whereas the DSL images were upscaled using "bicubic". No additional sharpening was done after scaling. I would normally sharpen after scaling but didn't do it this time to avoid adding parameters.


Some effort has been done to gray balance the the images and also to adjust the tone curve to match gray squares 1,3 and 6 on the "Color Checker" card.


Finally, all the results are shown in color and monochrome.


My own observations

To me it seems that the Velvia image is about on par with the DSL image when sharpened "correctly". The DSLR image is much easier to work with. One interesting observation is that the seems to be some paper structure in the Velvia image that is not visible in the DSL image. I may speculate that this is just film grain. Th DSLR image can take extreme sharpening with little degradiation. DR is definitively an issue with the Velvia, exposure scanner combination.


Native resolution for MF and Alpha upscaled to 12000 pixels width


Velvia 4800 PPI, no sharpeing Alpha 900, upscaled, no sharpening
Velvia 4800 PPI, sharpened Alpha 900, upscaled, sharpened




Velvia 4800 PPI, no sharpeing Alpha 900, upscaled, no sharpening
Velvia 4800 PPI, sharpened Alpha 900, upscaled, sharpened


These images have been scaled to 8000 pixel width:



Velvia , downscaled, no sharpeing Alpha 900, upscaled, no sharpening
Velvia 4800 PPI downscaled, sharpened Alpha 900, upscaled, sharpened



Black and White

Velvia 4800 PPI, downscaled, no sharpeing Alpha 900, upscaled, no sharpening
Velvia 4800 PPI, downscaled, sharpened Alpha 900, upscaled, sharpened


The sharpening game

One of the issues is that all digital image manipulation includes sharpening. Sharpening does not really improve resolution but plays a major role for the perception of sharpness. In the case the aberrations of the imaging chain are well known (that is the Point Spread Function is well known) actual resolution can be restored using a technique called deconvolution.


The problem is that different images need and allow for different amount of sharpening. Some scanners may even sharpen when sharpening is set to zero, negative sharpening may need to be set to avoid sharpening. Sharpening normally enhances edges and noise.


The most correct comparison can be probably be made with "correct sharpening", but this is hard to define.


Commented comments


NOTE! There has been an intensive discussion about the validity of this testing on Luminous Landscape with a lot of good (and also less good) points. The main idea was:

There is some truth to that. On the other hand:

Original images are here. You may use these images freely, they have been shot for these tests.

Comparing the images

Comparing the images is not trivial. To begin width both number of pixels and colors are quite different. Also, sharpening is an essential part of the processing chain. I have taken the following choices:

Comparing using Imatest

A comparison using Imatest was made. The results were

Camera LW/Ph
Pentax 67 + Velvia 2271
Sonya Alpha 2890

According to this the Sony is significantly better than the Pentax. This is not really consistent with my observations. I made really stunning 70x100 cm prints from Pentax slides, seems better than 2271 LW/PH to me. The measurement was made with default sharpening on i Vuescan, whereas the comparison iages were sharpened by Photokit Sharpener.


Checking feasibilty of Imatest results

I did a lot of testing using studio strobes and shooting subjects at different distance and moving the camera back and forth in small (3mm) steps. The best result I could achieve was about 2300 LW/PH. The lens seems to have quite visible chromatic aberration of axis. The finding of the imatest shoots is quite clearly that the Sony Alpha 900 full format DSLR with 24.5 MPixels delivers better performance than my Pentax 67/Velvia/Scanner combination using the 90/2.8 lens at f/8 and scanning at 3200 PPI. Next time I'll investigate the situation scanning with 4800 PPI.

Regarding the Imatest results I'd point out the following issues:

Looking at very small detail (in 300% size)

The pictures below are in native resolution, 3200 PPI for the scanned slide and actual pixels for the Alpha 900, both are shown in 300%. The text "ALARM" is easily recognisable on the scanned iamge but not in the digital image. The digital image is clearly pixelated. The probable explanation is that the slide has probable higher resolution on high contrast edges. On low contrast edges it may be difficult to see a similar resolution advantage for the slide.


The Pentax is known to have issues with vibration from the focal plane shutter and there was some concern about that affecting the resullts. As such small detail is clearly observable, I think we can rule out camera movement as a major factor in possible unsharpness on the scanned image. Further I actaully made measurements with Imatest in both horisontal and vertical directions the difference in the results was very small. As any vibration from the shutter would act horisontally a significant difference would be seen. There was some difference actually (about 9%), but this doesn't seem to be a major factor.


Visual comparison 1, unscaled images


Visual comparison 2 downscaled slide image

The slide image above was pretty unsharp, so I decided to scale it down to the same size as the digital image using Bicubic sharper. The images are now essentially same size.

Here the images are pretty close in sharpness. One area where digital seems to be better is tonal separation in red roses. Regarding the amount of detail it may be relatively little difference.


My estimate, for now, is that the Alpha 900 is probably good enough to fully replace my Pentax (which I nevertheless keep, it's a nice camera). I'll do some more checking, and try to sort out why it lacks performance in Imatest. Film has a long turnaround time, unfortunately, so some testing (like checking focusing accuracy) takes long time.

Planned tests

I plan some more tests. Primarily focusing accuracy on the Pentax 67. The present plan is to shoot a "Lens Align®" with the Pentax and also a series of test targets at slightly different distances on the same image. I also want to look more into other means of reproduction like projection of 67-slide and macro photography of the slides.


An area I also want to investigate is the separation in the red roses.


Other observations

The Pentax 67 is a rather decent medium format system for film but not the best one ever built. Mamyia 67 is known for it's fine lenses and a well built Rodenstock Digitar plays in another divsion. The Sony Alpha and the 24-70/2.8 ZA lens are probably as good as it gets, at least regarding sharpness. I won't argue that Nikon 3DX isn't better but the Sony is pretty good by today's standard.


Another way to see it

What the research on this pages really indicates that there is no great advantage of Velvia on 67 Medium Format over 24.6 MP digital. Neither has it proven that digital is superior to MF, at least regarding resolution. Still, I feel that the experience points in the direction that digital is more practical than MF film, for the following reasons:

The only situation I see where analogue MF can be of advantage is if one has a significant investment in analogue equipment (including scanning equipment), very good expertise in using it and small volume,



Bernard Languillier has been kind enough to point out that the Alpha 900 doesn't show the same pixel level sharpness he experiences with the Nikon 3DX which should have similar characteristics. I made a similar observation and did some checking and also compared with another shot. I cannot see any obvious problem. A similar problem is visible in DPReviews test of the Nikon 3DX:, they attribute the difference to weaker low pass filter on the Nikon, but it could be erranious focusing, bad lens or something else.



Last Updated on Sunday, 08 November 2009 19:45