I've figured out the difference between Merrill and Quattro

MarkWW

Leading Member
Messages
881
Solutions
1
Reaction score
246
Location
Brooklyn, NY, US
Cropped in the extreme out of deference to the model, who is quite lovely. We all know the Merrills are horrible at skin- they give a leathery look to even good looking skin.

We're starting from an OOC JPG with everything set to mid/default (sharpness, contrast, etc.) (anyone else have a problem with the DP3m where it says it's processing a file & keeps going for more than a minute? - well I had that happen to me, so I took out the battery and put it back in again & the camera was reset to JPG only, so I had no RAW files from a good portion of this shoot) that's been slightly processed in photoshop to adjust the color & contrast.

I was following along with a photoshop tutorial (on PHLEARN.com) that explains how to separate out the color data form the luminance data & then blur the color layer to even out skin tones, while still keeping texture. I thought - hey that sounds like a fun technique, let me try that. The results are below, a bit artificial looking, but it can be dialed back in photoshop.

Looking at the results I thought - wait, I've seen that look on skin before. A few weeks ago someone posted some photos of some surfers at the beach and I remember thinking that their skin looked simultaneously solid and textured, like a photoshopped Merrill.

And then it hit me. I'm blurring the color layer and adding in a luminance layer on top... This is Quattro.

Again, this is an extreme example with more blurring than the Quattro would do (and again I may dial it back before finalizing the image).

DP3m top, blurred color layer bottom.
DP3m top, blurred color layer bottom.

Steps to reproduce in photoshop.

1. Copy your base layer & set a gaussian blur on it - not too much, just so that the details get smeared. Let's call this layer "Color".

2. Copy your base layer again and put it above the color layer. Now go to Image, Apply Image. Set the layer to the Color layer, the Blending mode to Subtract and the numbers below at 100%, 2, 128. Once you're done, set the blending mode to Linear Light. Let's call this layer Luma.

You now have a duplicate of your image with the colors blurred & the luma separated out. It should look identical to your original image. You've basically created a low pass layer (Color) and a hi pass layer (Luma), so it's not exactly the same as a Quattro, but the results are similar.

Right now you have a copy of your original image with two sets of data - color & luminance (or something approximating them). The next step is what makes it look like Quattro.

There's an action on the Phlearn website for this if this is too many steps to follow along.

3. Apply more blur to the bottom layer - either in patches (as you might for skin smoothing) or if you like, to the whole image. Again, just enough blur so that the details you wish you to smear get smeared.

What are you left with? It's Quattro.

This is a more extreme example with a lot more color blending going on than the Quattro does, but try it on a couple of your Merrill photos and let me know if you think the results look like Quattro.

Now I'm going to get back to editing this pic.
 
Very interesting, You certainly have given it a bit of thought.

Cheers.
 
Cropped in the extreme out of deference to the model, who is quite lovely. We all know the Merrills are horrible at skin- they give a leathery look to even good looking skin.

I was following along with a photoshop tutorial (on PHLEARN.com) that explains how to separate out the color data [from] the luminance data & then blur the color layer to even out skin tones, while still keeping texture. I thought - hey that sounds like a fun technique, let me try that. The results are below, a bit artificial looking, but it can be dialed back in photoshop.
Interesting post, Mark, should garner a lot of PO ;-)

Did the processing involve the L*a*b* color space?
 
Cropped in the extreme out of deference to the model, who is quite lovely. We all know the Merrills are horrible at skin- they give a leathery look to even good looking skin.

I was following along with a photoshop tutorial (on PHLEARN.com) that explains how to separate out the color data [from] the luminance data & then blur the color layer to even out skin tones, while still keeping texture. I thought - hey that sounds like a fun technique, let me try that. The results are below, a bit artificial looking, but it can be dialed back in photoshop.
Interesting post, Mark, should garner a lot of PO ;-)

Did the processing involve the L*a*b* color space?
 
Cropped in the extreme out of deference to the model, who is quite lovely. We all know the Merrills are horrible at skin- they give a leathery look to even good looking skin.

I was following along with a photoshop tutorial (on PHLEARN.com) that explains how to separate out the color data [from] the luminance data & then blur the color layer to even out skin tones, while still keeping texture. I thought - hey that sounds like a fun technique, let me try that. The results are below, a bit artificial looking, but it can be dialed back in photoshop.
Interesting post, Mark, should garner a lot of PO ;-)

Did the processing involve the L*a*b* color space?

--
Cheers,
Ted
No, I never really bother going into LAB color space. Photoshop is one of those infinite beasts where everyone has their own process for working. I'm familiar with LAB color space, but only use it on the rare occasions (as in once every few years) where it does something better than my usual method of working.

I basically just created a hi pass layer (detail) and a low pass layer (color) and blurred the color layer.
Thanks Mark, now it is clear. I only have Elements and hardly ever use layers - so I wasn't following the description too well.

Along the same lines, I did something related, some time back. Took the red and green channels, downsized them 50% with the 'bicubic' selection. Re-sized them back up to 100% but with the 'nearest neighbor' selection. Then merged those two channels with the untouched blue channel in ImageJ. Result was very similar to yours!

--
Cheers,
Ted
 
Last edited:
I am sceptic

It looks like a lot of blurring.

What would happen if you did that to a thin red thread with a background of another color? My guess is that the red thread will lose its color totally. Is that how Quattro handles that case?

Moreover, I think the color smoothed skin also looked strange. Like she is heavily freezing.
 
Last edited:
I am sceptic

It looks like a lot of blurring.

What would happen if you did that to a thin red thread with a background of another color? My guess is that the red thread will lose its color totally. Is that how Quattro handles that case?

Moreover, I think the color smoothed skin also looked strange. Like she is heavily freezing.

--
/Roland
X3F tools:
http://www.proxel.se/x3f.html
https://github.com/rolkar/x3f
Clearly this isn't EXACTLY what's going on - the hipass layer that I create with this technique is based a blurred lower layer, the blurred lower layer has edge bleeding & that gives the hipass layer halos, and as I said before, this is far greater amounts of blur than Foveon would apply.

Additionally, I think the skin texture on the Quattro sample pics I saw a couple weeks back looked strange as well - or at least, different from any Bayer or Merrill photo of people I'd ever seen.

That said, despite the heavy amounts of blurring, if you add the original layer back in as a luminosity layer on top - et voila, it's very much like the original Merrill image.

I did a similar experiment last week where I created a color & luminosity separated image following these intructions (with some tweaks), then downsampled & upsampled the color layer while keeping the luminosity layer intact and it looked identical to the original. Even zooming in and pixel peeping I couldn't tell the difference.

Top Left - original, top right down/upsampled, bottom left luminance data, bottom right luminance applied to the upper right color data. Upsized 800% using nearest neighbor.
Top Left - original, top right down/upsampled, bottom left luminance data, bottom right luminance applied to the upper right color data. Upsized 800% using nearest neighbor.

Blurring further (downsampling & upsampling further) lead to loss of color fidelity as colors blurred together, but the sense of sharpness was the same.

And yet, Quattro images have a different look to them than Merrill images. Edges are sharp, but surfaces are smooth, almost like a sort of spherical aberration.

Which leads me to think that the Quattro isn't JUST upsampling the lower layers and applying luminosity from the top "panchromatic" layer.

Perhaps there's some edge detection going on in the decision of when & how to apply the panchromatic layer. Not literal edge detection, but the the processing engine sees a 2x2 cluster of sensels that in the bottom two layers are the same, but a major change in the top layer and applies the top layer more strongly at this point, thus preserving the edges.

Meanwhile on surfaces, the differences aren't that great, so the processing engine applies less of the top layer - just enough to give some definition to each individual pixel.

So the processing engine sees this (using values from 0 to 9).

RR GG BB
RR GG BB (or B/M/T or whatever).

33 55 78
33 55 67

It just sort of average the differences out a bit, giving something higher definition than you'd get from the bottom and middle by gently applying some of that top layer value to the lower layers.

But then it hits an edge

33 55 29
33 55 38

The values on the top layer are very different - time to apply more of the luminance layer to the lower layers.

I'm not a whiz at math, but something vaguely along these lines could cause the Quattro look of sharp edges but smooth surfaces.
 
A sort of process of mummification...or cryopreservation?
 
Which leads me to think that the Quattro isn't JUST upsampling the lower layers and applying luminosity from the top "panchromatic" layer.
+1 to that, especially the latter part . .

Much discussion about that here:


and here:

 
Tell Lindsay that the Merrill sensor makes skin look leathery. She shoots with it all the time. Here's her Web site, so you can see some of her work with beautiful models.


Leathery . . . lol.
 
Tell Lindsay that the Merrill sensor makes skin look leathery. Leathery . . . lol.
It does, but maybe photoshopped models with perfect make up in perfect light aren't perfect to proof the contrary.
But against leathery skin you also can use iridient developer or the SPP monochrome mode for the luminance/rendering.
 
Have you tried just decreasing fill light and compensating by adding exposure to an x3f raw.

eg:







--
'Everything in photography boils down to what's sharp and what's fuzzy.'
-Gaylord Herron
 
Tell Lindsay that the Merrill sensor makes skin look leathery. She shoots with it all the time. Here's her Web site, so you can see some of her work with beautiful models.

http://www.lindsayadlerphotography.com/#!/index

Leathery . . . lol.
Unfortunately, Scott, the link takes us straight to the Golden Goddess whose skin is about as leathery as it gets (look at the hand) and then in Lindsay's portfolio we find an Ice Maiden with no skin pores or facial hair at all!

But I do agree with you that it's not the sensor - it's all in the processing . .
 
Cropped in the extreme out of deference to the model, who is quite lovely. We all know the Merrills are horrible at skin- they give a leathery look to even good looking skin.
No, it has a bit more micro-contrast.
<...>

I was following along with a photoshop tutorial (on PHLEARN.com) that explains how to separate out the color data form the luminance data & then blur the color layer to even out skin tones, while still keeping texture. I thought - hey that sounds like a fun technique, let me try that. The results are below, a bit artificial looking, but it can be dialed back in photoshop.

Looking at the results I thought - wait, I've seen that look on skin before. A few weeks ago someone posted some photos of some surfers at the beach and I remember thinking that their skin looked simultaneously solid and textured, like a photoshopped Merrill.

And then it hit me. I'm blurring the color layer and adding in a luminance layer on top... This is Quattro.
This is totally wrong. Skin on the Quattro looks NOTHING LIKE THAT. It's simply not as sharpened as a Merrill image. It still requires skin smoothing software in most cases... It still maintains fine detail o the skin like pores and other things the technique you outlined obliterate.
Again, this is an extreme example with more blurring than the Quattro would do (and again I may dial it back before finalizing the image).

DP3m top, blurred color layer bottom.
DP3m top, blurred color layer bottom.

S
--
---> Kendall
 
not as extreme as what you are showing (which you stated).

But I would call it a step in that direction.
 
Have you tried just decreasing fill light and compensating by adding exposure to an x3f raw.

eg:



--
'Everything in photography boils down to what's sharp and what's fuzzy.'
-Gaylord Herron
The skin looks good, though it changes the tonality of the whole photo.
 
Tell Lindsay that the Merrill sensor makes skin look leathery. She shoots with it all the time. Here's her Web site, so you can see some of her work with beautiful models.

http://www.lindsayadlerphotography.com/#!/index

Leathery . . . lol.
I accept your statement, but unless you know the model & makeup artists she works with, her lighting technique and post processing workflow, you shouldn't use her as an example to counterbalance a statement I'm making based on my own personal experience with the DP3m.

Below is a photo - left is the TIFF straight out of SPP, right is from my final edit in photoshop - I could probably have pushed it more, but this was already my "extreme" edit and I didn't want it to look "high fashion" - I wanted to keep it a bit more in the realm of realism.

Thanks to a combination of the model, makeup and photography (lighting, camera etc.) even the 2nd image feels somewhat "unreal" - it's not what we're used to seeing in photos, nor is it quite what we're used to seeing in person either.

So when I say skin looks leathery with the DP3m, you must understand I don't mean all people, all the time or that it's not something that can't be controlled. I was simply going through my thought process in attempting this new photoshop technique I learned online (not used at all in the image below).

BTW I'm hoping that the Quattro will treat skin more like on the right (again, needs good model + good lighting + good makeup to look like that) but basically "Foveon without the micro contrast."

DP3m. Left: Straight out of SPP, right with some photoshop.
DP3m. Left: Straight out of SPP, right with some photoshop.
 
Last edited:
Tell Lindsay that the Merrill sensor makes skin look leathery. She shoots with it all the time. Here's her Web site, so you can see some of her work with beautiful models.

http://www.lindsayadlerphotography.com/#!/index

Leathery . . . lol.
Unfortunately, Scott, the link takes us straight to the Golden Goddess whose skin is about as leathery as it gets (look at the hand) and then in Lindsay's portfolio we find an Ice Maiden with no skin pores or facial hair at all!

But I do agree with you that it's not the sensor - it's all in the processing . .
 
I agree, if you're saying the micro-contrast is "toned down a bit" . . . but I think the micro contrast is still there - just not as harsh. It can be enhanced in Photoshop, from what I'm seeing in people's demonstrations. Will a Quattro image ever look as full of micro-contrast as a Merrill image? Maybe not. But ultimately that would be a good thing for Sigma, if they decide to sell two lines of cameras (one with Quattro sensors and one with Merrill sensors).
 

Keyboard shortcuts

Back
Top