Sigma SD1 review

[Graeme Nattress]

"I get the impression, but correct me if I'm wrong, that the Foveon design not only has the benefit of no AA filter, but the benefit of 'no need' to interpolate (or invent) image information because of the unequal balance between the number of red, green and blue diodes that the Bayer system has to contend with."

Benefit or curse though? Depends what you're shooting and your personal aesthetic preference or not for aliasing. The lack of OLPF is not a Foveon thing per se, as Michael had the good sense to round up some OLPF-less Bayer CFA cameras for comparison. If you want no OLPF, but a compact non-MF form factor there are choices to be had.

As for the lack of interpolation - there is no lack of interpolation. The chroma all has heavy NR applied which is absolutely inventing new (less noisy) image data. Indeed, the image processing necessary to turn a raw Foveon image to a nice RGB image is more complex and more intensive than is used to turn a Bayer mosaic image to RGB. I'd contend there's more interpolation going on for Foveon images than for Bayer. The argument that in Foveon there's no interpolation is false.

Graeme

[hjulenissen]

I get the impression, but correct me if I'm wrong, that the Foven design not only has the benefit of no AA filter, but the benefit of 'no need' to interpolate (or invent) image information because of the unequal balance between the number of red, green and blue diodes that the Bayer system has to contend with.

The 'qualitative' benefits of the Foveon system in this respect, disregarding pixel count, should exceed the qualitative benefits of MFDB.

People talking about the need to "invent" or "interpolate" color information in Bayer sensors vs the "true" pixels in Foveon sensors tend to rely on home-grown (or PR-driven?) theories that tend to be less relevant and obscure the understanding rather than enlighten. Not saying that you are, just explaining why "interpolated Bayer pixels" tends to set off my mental warning lamps. Even though I use theories every day, I really think that the IQ of image sensors should be finally judged by a sound collection of representative side-by-side test images judged by our eyes.

The AA-filter used in cameras and CD-recorders is a cost-adding component that is added for a reason. Canon and Sony does not add to their costs unless they think that it will increase sales. Some might disagree and prefer the unfiltered sampling, but that is a subjective thing. The basic theory behind was well understood when I was in University some years ago.

-h

[bjanes]

Where did you learn that SD1 is gong to get the DxOMark treatment? I couldn't find such information on their site, but didn't really dig deep If true, it sounds interesting.

Look here.

Of course, DXO does not test resolution so we will have to look elsewhere for those results.

Regards,

Bill

[Ray]

Benefit or curse though? Depends what you're shooting and your personal aesthetic preference or not for aliasing.

Exactly! However, what some people may recognise as aliasing, or false detail, other less expert viewers may find pleasing, even startling in its crispness.

As for the lack of interpolation - there is no lack of interpolation.

There is a lack of interpolation in the sense that there are real values of red, blue and green for each pixel as opposed to two values of green for each pair of red and blue in the Bayer array, and the description of any one of those values as a whole pixel.

To put it another way, a 15mp Bayer array sensor gathers 15 million separate values of red green and blue. A 15mp Foveon sensor gathers 45 million separate values of red, green and blue. That's not an insignificant difference.

Of course I understand there are some diadvantages related to the absorption characteristics of the silicon. It's not an ideal color filter, and high ISO is rather noisy in the Foveon sensor compared with the best Bayer array sensors, but MFDB is also notorious for poor performance at high ISO. That doesn't stop MFDB owners from drooling over the extra crispness and intangible 3-dimensionality of their images.

I'll always remember my amazement when I first learned, in the days when I was still shooting film, that a 3mp digital camera has only 3 million separate color values, even though the image file ends up being 9MB. That's a hell of a lot of interpolation, I thought. In fact, as I recall, impressions of cheating and dishonesty passed through my mind. I remember being disappointed.

[deejjjaaaa]

The default profile for the SD14 is described as "Embedded".

exactly - it is hardcoded in the code

I believe the earlier versions of LR must not have supported DNG for Foveon (I read a support forum post to this effect from Thomas Knoll) but if definitely does now.

yes, only since ACR 6.3 RC 

[Graeme Nattress]

Ray, what does it matter if you start off with a unique value for each of RGB at a single location if you then don't pass those unique values through to the final image because they're too noisy to use as-is? Noise Reduction is an interpolation process where we try to figure out what the under-lying noise free signal is from (spatially) multiple noisy samples. Figuring out a new value from a spatially diverse set of samples is exactly what you do in a demosaic process too. Both are interpolation, and "interpolation" is not a bad word in my book, although it is often used pejoratively. Almost everything we do in image processing to generate a pleasing image from raw data is some form of interpolation. In many demosaic algorithms the existing samples are passed though as is, meaning 1/3 of the final RGB values have not been interpolated. In the Foveon raw process, (if we assume for the moment, giving them the benefit of the doubt) the chroma pixels are all noise reduced and hence are not the originally captured values, so 2/3s of the final RGB are interpolated - leaving 1/3 that have not been.

The process of turning a Foveon raw into an image is at least as "inventive" as what goes on in a Bayer demosaicing process - you're just trading more precise, but lower spatial chroma data for higher amount of lower precision chroma data.

I must admit though, I find the "Foveon" name funny, because the fovea of the eye uses a spatial array for the LMS cones more like a Bayer pattern than a layered film-like approach.

Graeme

[Sareesh Sudhakaran]

A combination of the Foveon technology in a Bayer pattern matrix will be a phenomenal beast, wouldn't it? Now why didn't Sigma think of that?

[Ray]

Ray, what does it matter if you start off with a unique value for each of RGB at a single location if you then don't pass those unique values through to the final image because they're too noisy to use as-is?
Graeme

Graeme,
The results indicate it matters a lot. A 15mp Foveon sensor provides more detail, or at least an impression of more detail and better micro-contrast, than any existing cropped-format DSLR on the market. According to Michael's article, the SD1 provides marginally more detail than the 24mp Sony A900.

Some of the images from an SD1 owner at Dpreview look stunningly sharp to me. I'm reminded of certain comparisons between MFDB and 35mm DSLRs that used to be the source of endless debate on this forum, except the differences from the SD1 seem to be greater in terms of micro-contrast.

[Graeme Nattress]

Ray, so it sounds like you like the look of no OLPF. If colour accuracy is not an issue for you, and you generally shoot at low ISO where noise is less of an issue, then great. Other thing you can do is shoot with a higher resolution Bayer CFA camera and downsample it -  that can induce the same kind of lack-of-OLPF look you like (depending on choice of downsampling filter), could also lead to still lower noise.

Sudhakaran, the problem with Foveon is is not that it samples RGB at the same spatial point, but that to enable it's sampling of RGB at the same point it uses silicon thickness as a colour filter and that leads to poor and noisy colour. With a Bayer CFA you get good colorimetry and lower noise, but you sacrifice the layered arrangement. With Bayer CFA you can always produce a sharper image and the appearance of spatially co-located colour through downsampling. With Foveon there's very little that can be done about the colour.

Graeme

[John R Smith]

Supposing - just supposing - that you didn't actually want colour at all, but black and white.

Which sensor type would give you the best starting point for B/W conversion - Beyer, or Foveon?

John

[Graeme Nattress]

John, I don't think there's a simple answer there. I'd always said we should have answered the eternal question: "What's black and white and red all over?" with a RED camera without the colour filters on the sensor. Then you'd use colour filters just as you'd shoot with black and white film. That would be the lowest noise solution.

With a Bayer CFA, you can post-choose your colour filter from careful blending of the RGB channels. This makes for good results and Bayer CFA cameras tend to have better noise and ISO performance - better dynamic range.

With a Foveon by shooting for black and white, you throw away it's one advantage, it's co-sited colour, but you get it's noise and ISO performance disadvantages.

Graeme

[kwalsh]

Regarding what Graeme has been saying already, noise filtering and interpolation are actually identical processes.  Both involve spatial filtering.  This often isn't obvious to the casual reader, but the reality is the signal processing is equivalent.  Ray appears to be conflating "interpolation" with "unsharp" and there really isn't justification for that.  Yes, the Foveon imager produces very high resolution for its pixel density but that isn't necessarily because of "interpolation".  Both Bayer and Foveon images have spatial filtering applied.  As has been stated already, an advantage of the Foveon in regards to spatial luminance resolution is the lack of an OLPF.

As far as B&W goes I've wondered the same.  I was anxious to try that out with an SD1, but since it looks like no one is actually going to own one I guess I won't.  Interesting to note you can get a few Canon models modified to remove the CFA and OLPF for what should be extra sharp B&W images.  I've asked more than one place about results with these cameras but have never found someone actually using one!

Final note on the SD1, I was given to understand this imager does in fact do CDS using a different pixel cell than the SD15.  Probably the reason for the improved ISO performance.

Ken

[Rob C]

Fred, I'm not sure if Robin Wood was intended, or lost in translation or a typo. But whatever, it's a great line made all the better!

That's my boy Fred! Once you've tasted Jurançon you have that 'otherness' the rest can only envy.
;-)

Rob C

[Aku Ankka]

Look here.

Of course, DXO does not test resolution so we will have to look elsewhere for those results.

Regards,

Bill

Hi Bill,

DxO also has similar place holders for three other Sigma cameras, yet no reviews, nor did I find anything on the link you provided mentioning of a future review.

[fredjeang]

I used Robin Wood as an image, sort of saying "the one who saves the poors".

I was in Nothingham in my youth, there is really the famous tree. Don't know if it's fake or Robin Wood actually used it or if all that is a legend like the scotish lake's monster (ups...I should not say that, there are scotish photographers around here  ). Really loved Nothingham city, very elegant.

Cheers.

[bjanes]

Hi Bill,

DxO also has similar place holders for three other Sigma cameras, yet no reviews, nor did I find anything on the link you provided mentioning of a future review.

Look at the top just above specifications where it says: "This product will be tested and reviewed on July on dxomark.com. Stay tuned by subscribing to our newsletter. " July is almost over, so they might be a bit behind.

Bill

[fredjeang]

Fred

Robin Wood sounds so much more poetic than Robin Hood, but either way, that's still one of the best lines I've read ;-)

  oh dear...I did not see my mistake! Do you know why? it's because in french Robin Hood is Robin des bois , and bois means wood.

Ok, if more poetic I stay with Robin Wood then.

Cheers.

[Rob C]

oh dear...I did not see my mistake! Do you know why? it's because in french Robin Hood is Robin des bois , and bois means wood.

Ok, if more poetic I stay with Robin Wood then.Cheers.

Please do; it's far more cool!

I also believe that I have some secret mojo working for me when I speak Spanish here; I can always find what I want if it costs money but it gets strangely difficult when it should be for free. Hmmm....... must be Scottish charm working backwards - a time-zone slip or something similar.

Rob C

[hjulenissen]

With a Foveon by shooting for black and white, you throw away it's one advantage, it's co-sited colour, but you get it's noise and ISO performance disadvantages.

Spatially, a Foveon used for B&W images (summed color channels) is equivalent to a true monochrome sensor.

Co-sited subpixels (Foveon) means that the need for OLPF is reduced (Sigma leaves it out altogether). This means that an APS-C sensor featuring 15 unique mega sensel locations chould provide high-acutance "B&W" images (somewhat aliased).

A Bayer APS-C sensor with no OLPF of 15 unique mega sensel locations might provide similar acutance for B&W (and better noise/ISO characteristics, lower price etc) but should have more pronounced problems with aliasing (even when used for B&W), and I am unaware of any such sensors available in off-the-shelf cameras.

I think that the situation is complicated by the fact that accurate capture of color is good even for B&W photographers, as it is used for post-processing simulation of color filters for creative effects.

-h

[Graeme Nattress]

The need for OLPF is only reduced with respect to to chroma moire, not aliasing in general to which it is as prone as any sampled system. I'd be keen to see how the Foveon performs as a black and white sensor because if you're not decoding colour from it, you're not getting into issues of matrix induced noise. It could work well, unless the extra circuitry in there to get the colour is detrimental to dynamic range.

Given the availability of Bayer CFA sensors with low noise and higher measured resolutions (with lower aliasing) than 15mp, I'd probably go that route myself, use post processing to generate the BW from colour and if I wanted that crunchy aliased look just use a crap downsample filter. I'm sure I'd save myself money that way so I could spend on more good glass.

Graeme