« Man Seeking Woman | Main | John Stanmeyer Wins World Press Photo of the Year »

Friday, 14 February 2014


Feed You can follow this conversation by subscribing to the comment feed for this post.

I confess when I saw this exotic beast my first thought was that Sigma must be VERY confident in the image quality to put out such a 'radical' looking camera. I mean, when you go through life looking like that you just have to be able to punch well above your weight, don't you?

It's a weird beast, no mistake. There's a bigger battery (looks about twice as big in the shots I've seen) which addresses one of the big problems of the DP series. The whole idea of the sensor where the blue layer has four times the resolution of the red and green layers is weird. But I want one.

I bought one of the original DP1 cameras when they first came out. In many ways it's a terrible camera. ISO range is limited. Autofocus is slow, saving to memory card is very slow, video is a joke they added just so they could tick it off on the features list.

But i love the images I get out of it, which is why I've kept it.

I'm looking forward to seeing samples from this.
Or maybe I'll just look around for a cheap DP2Merrill and buy lots of spare batteries...!

Personally I don't only think it looks foul, but it seems to be the most inconvenient shape it could be. Just because you can, doesn't mean you should - especially when the world seems to be shouting 'we want cameras that look like cameras'. I can't quite work out why the previous design had to be ditched so entirely. In a word, 'no'. In many senses.

I have a dp2 Merrill. When I use it within its limitations (low ISO, hold the cam steady, stationary subjects) it can produce remarkably detailed images with luscious colour that can take my breath away. Very occasionally the colours are a bit "off" but mostly not. The files are huge and the tiny 350 gram camera struggles to process them rapidly. It takes a lot of juice to process the images so the cam goes through batteries very quickly. I always carry several. Other than eating through batteries the camera doesn't do quick. Because of the way it records luminous info black and white images are impressively rich and detailed. The proprietary file format means you have to endure passing the images through Sigma's hellishly slow and arcane proprietary software to batch crank out 16 bit TIFFs for Lightroom et. al. The lenses on the various Merrills are limited to f2.8 but are excellent.

This new cam in large part was designed to ameliorate many of its predessors faults. It uses a reworking of traditional Fovean architecture which theoretically has promise to improve the already superlative image quality of the dp Merrills while reducing file sizes hence recording time in camera and processing time in desktop software. ISO may improve 2 stops making 1600 the absolute upper limit for colour and maybe a stop more in b/w. It now records 14 bit files which may help in retaining yumminess. The grip handle houses a significantly larger battery. Lenses remain the same.

Of course the proof of the pudding is in the eating so no amount of analytic speculation will tell us what we each might want to know about any cam. The massively quirky dp Merrill wasn't a camera most people would want and I suspect the same will be true for the Quattro but if it's an improvement over the Merrill it should be an engaging image making device. Personally I'm curious to hold one. If the ergonomics work and the files are good the Quattro will make a nice walkabout MF like camera.

The "strange, futuristic, and somehow oddly appealing shape of the camera."

I'm pleased you call it appealing. I've always seen the shape of the DP series as appealing. These new Quattro ones are certainly unusual, but I still think they're good looking. I'm a lot more skeptical of the practical merits of the design, but I hope that skepticism is mistaken.

I wouldn't take anything Engadget says about photography seriously.

Apologies for this being so long. It will be tl;dr for most but the devil is in the details. All IMHO as I've not made any measurements on a Foveon sensor.

The Sigma still claim that the three layers are blue, green and red. They even do this in their patents. But it's just not true.

The layers are separate but they don't just detect R, G or B each detects a mix of the primaries. Some blue light is absorbed in the top, middle and bottom layer. Some green light is absorbed in the top, middle and bottom layer. Some red light is absorbed in the top, middle and bottom layers. Without dyes to form a definite color cutoff (like Bayer) you are relying on absorption in the silicon to do the filtering which is gradual and probabilistic.

The probability of a photon being absorbed in the top layer is highest for the blue, next most light for greenish and least likely for the red but all three "color ranges" do get some absorption in that layer. It's a bluish cyan.

In the middle layer some of the blue has been filtered (but not all) so that middle layer is a desaturated yellowish green (some blue, less red and mostly green).

The bottom layer is orangish red (mostly red, a little green and a very little blue).

To get to real RGB primaries you have to process these (noisy) "raw" color signals though a color matrix to remove the common-mode crosstalk between the colors giving even noisier RGB signal out as you are multiplying and subtracting three noisy signal to get to a R, G or B primary.

You can see the response of the original Foveon sensor plotted (in their patent US07132724).


This is the major reason for Foeveon sensors poor high ISO performance. The signal to noise ratio goes down quicker than you'd expect compared to other sensors. It also contributes to poor color fidelity because the separation of color is much poorer than dye based primaries with sharper cut off and a closer match to human color receptors.

"White light" is about 60% greenish, 30% reddish and 10% bluish in terms of photo counts so even though the top layer is most sensitive to blue there is a lot of green (when taking real images) absorbed in that layer. That's the reason they can subdivide the top layer into four subpixels or consolidate the middle and bottom layer into 4x larger pixels (depending on your POV).

The other problem with Foveon sensors is the pixel is big compared to single layer ("Bayer") sensors because you need extra silicon to bring the charge up from the buried PDs. Recall the last Foveon sensor had only 5.4 megapixels (each recording 3 colors). This causes a problem in the "pixel wars" and a problem for Sigma marketing.

The new "Quattro" sensor design was worked is credited to the late Richard B. Merrill by in US patents 8039916 and 7834411 based on work done in 2007 (and filed in 2010) to work around both of these problems.

This "Quattro" design works by reducing the resolution of the bottom two layers so as to improve their noise performance and so overall noise performance whilst trading off color resolution at the red end of the spectrum. This is an admission that the top layer isn't really "blue" otherwise you wouldn't get most the luminance information at high resolution (which is mostly in the green channel that's why Bayer CFA have two green pixels). The top layer clearly has a lot of green (luminance) data otherwise this reduced resolution in the lower layers wouldn't work.

All the subpixels share a single (fairly standard) 4T-like APS design (but there are more transistors to switch each PD to the sensing node). Even with all this sharing there is still a fair amount of silicon dedicated to moving charge up from the lower layers.

So now Foveon has all high photo noise issues (from subtracting all three channels from each other) along with a variation of the demosaicing problem (they have to do color interpolation to get RGB for each pixel). This seems to be the worst of both worlds.

I suspect they don't do any interpolation (demoasicing) when they generate RGB but just use the value of the appropriate "big" middle and bottom later pixels (divided by 4) in the usual matrix method. I suspect they just live with the reduced spatial resolution of the middle and bottom later and so reduced resolution in red and yellow.

There also the problem of the pixel count in marketing. This is a 19Mpx resolution sensor in greenish cyan (and perhaps most of the luminance channel) and a 5Mpx sensor in the greenish yellow and organish red layers. Yet they claim it's a 39Mpx (eq) sensor. Bizzare. We just add the count of the PDs together. . At best (blues and greens e.g. landscape photography involving leaves?) it's a 19Mpx sensor but for other colors the resolution will be worse but perhaps no worse than Bayer.

I suspect the Foveon sensor idea may have come to the end of the line with this design. They've had to give up the idea of all three layers having the same resolution which seems to be the original core idea. The trade off for no color moire (but not no luminance moire) is worse color performance and worse high ISO performance and poor software support in popular products.

It seems that the Bayer CFA trade offs are not so bad: like democracy it's the "least worst" solution. Something for XTrans fans to keep in mind :-)

When I first saw a picture of this, I thought of a bent Linhof panoramic camera.

It's neat when some corporate entity tries out something way out there. I'd like to meet and speak with the folks on the design team, find out what they were thinking. I figure that even if it fails in the consumer marketplace, collectors may want one just for the hell of it.

Looks more like a phone than a camera.

It won't be long when Sigma fit all three lenses to that body…just stretch it a little bit more!

It will be interesting to see how well the sensor works in practice.

BTW, it's misleading to call the layers 'blue', 'green' and 'red', as the top layer catches a significant amount of photons in the 'green' wavelengths and a lesser amount of the 'red', along with the 'blue'.

The lower layer in the previous DP Merrill camera sensors had such a poor signal to noise ratio that they probably don't lose much in terms of colour resolution in the new scheme, and are likely to benefit greatly from reduced chroma noise.

I quite like the camera shape. It seems designed to mandate either a two handed grip, or use of a tripod - definitely not a camera for casual use.

It looks to me like a version of a 1998 era Nokia cell phone I had. With a lens attached. If it does have a good sensor, it will test whether those who say that only the results matter are sincere or not.
I, however, will never again say ONLY the results matter.

Koni Omega took Argus C3 to the prom and this was the
result. Dear God almighty! What an ugly looking baby. :)

Sigma was the first one that came up with the very appealing idea of big sensor, small camera. I still have the original DP1. It has been repaired once but still takes very nice pictures. But this is no longer small.
All these years Sigma has been highlighting the Foveon benefit that it does not need to interpolate color data:
'This three-layer full color capture system can record the full color information at each pixel location... without the compromises of color interpolation required by conventional sensors.'
Now it looks like they have to interpolate color to get the full pixel resolution. I am sure it is good, but so is a good Bayer sensor. And the sharpness destroying anti alias filter has been removed by competitors long ago. So where is the real world benefit now?

I've seen it referred to as Sigma DP Stretch. To me, it looks like my old 110 Instamatic. Just pop a flash bar on that baby.

It's great that the designers finally made room for a viewfinder. It's not great that they then failed to put the finder in.

I'll join those who find this camera appealing (design-wise). I even dare to advance the standpoint that it beats, ergonomically, the traditional design.
1. You have a substantial grip by redesigning the angle - but the function is the same.
2. When taking camera to the eye, your right hand is a bit further from your head than in the traditional design, which is quite convenient for those wearing glasses.
3. You have enough place inside the grip for a bigger battery.
Just as a curiosity: as there are electric guitars for left-hand players, why aren't cameras for left-eyed users. My left eye is much better than the right one. I'm lefty. I'd love this camera to have the grip on the opposite side, so I can hold it with my left hand, and lift it to my left eye, if an EVF is used. With Dslrs is not such a big deal, since you use both hands, anyway, and the viewfinder is on the center of the camera. But such compacts as Sigma's really beg the question: why not lefty cameras?

There is some explanation of the design of the chip in bythom.com once he got passed the first reaction to it. It make sense just not in the way Sigma put it.

I'd like to see a video of someone using this thing. There's a video on DP Review of someone handling it, but I want to see how it fits in the hand when using it.

Looks ok to me - at least it's seems big enough to hold with normal sized hands. Again, no viewfinder - EVF or otherwise! What's with this trend? The viewfinder is my "office".

The only real gain from the sensor will be 1.33 EV ISO benefit. Coupled with a couple of other changes Sigma is making, maybe you'll see 2 EV ISO benefit. So, instead of an ISO 400 camera max, you'll have an ISO 800 camera max.

The original Foveon design has dispersion in the lower layers (diffraction, focus, and reflections), meaning that they don't really contribute to resolution. Indeed, many Foveon users have discovered that blurring the bottom two layers nets them the same thing that Sigma is apparently now going to offer: same resolution, better low ISO ability.

Apparently Foveon themselves considered this design originally. Not sure why they didn't do it. I suppose there was some ego involved around the notion that you were getting RGB at every pixel. But you weren't. You were getting a color biased luminance up top, and two color differential channels below (yellowish and magentaish, something akin to a skewed Color Lab). The graphic that is on dpreview is deceptive, as it describes blue, green, and red layers. Think of it more as luminance, chroma difference, chroma difference.

Great, I've seen the files the Merrill produces and they are wonderful, but the package as a whole never appealed. I've got big hands and tired arms (from my full frame DSLR). So if it holds well in the hand it will be a serious contender. No more "finger clicking good" hand holding my compact.

There's a rather nice interview with the Sigma CEO over at dpreview in which he talks about the Quattro:

Thinking about it, the form kind of makes sense, as if you're holding a viewfinderless camera out in front of your body, it's probably easier to do so with your hands apart rather than close together.

What I liked most, though, was this:
My first priority is to make sure that the business continues to develop as a going concern, to protect our employees. We don’t need to grow enormously, we just need a slight growth to continue the business. Assuming we're doing OK, I’d like to do something amazing. Something to make customers say ‘wow’. That’s my real motivation. I’d rather do that than make our company bigger.

These are niche cameras, but I think every photographer owes it to themselves to experience the qualities of a Foveon sensor (try it via Amazon or a friend if you can). It really does have a fundamentally different look than regular Bayer sensors. I started with the DP1 when they became more affordable about 3 years ago. Then went for a DP2 for the focal length. I bought the DP2 Merrill a few months ago, a discounted refurbished model. For most people it won't be their only camera (or even their main camera); I have a Canon 5D2 as my main go to camera, as well as a Sony RX100 pocket camera. But the look and detail the Merrill is able to capture remains stunning. The Luminous Landscape has a thread demonstrating the potential of the camera.

Some interesting comments from


Has high ISO performance been improved?

It has been, yes. By around one stop.

So they must have improved QE a little bit (perhaps by improving the fill factor) and/or reducing read noise.

I see that just having a lot of sensels gives them problems of processing them (e.g. scaling, noise reduction and doing a 3x3 matrix multiply on all of them to get to RGB is quite expensive).

And I thought the Sony NEXs were ugly...

The comments to this entry are closed.



Blog powered by Typepad
Member since 06/2007