Four raw converters , handling demosaic errors

[ErikKaffehr]

Hi,

I converted an image from my P45+ using four different raw converters. The image was chosen as it had a lot of aliasing artefacts. What I wanted to see was if any of these raw converters handled the problem areas well.

The raw images is here: http://echophoto.dnsalias.net/ekr/Articles/RawConverters/Raws/CF045470.IIQ

The two attached crops are taken from different points of the image. They have been resized in Photoshop to 200% using bicubic interpolation for comfortable viewing. I tried to process the images as close to default as reasonable, but turned of sharpening off.

What I see is that Lightroom conversion is pretty bad on the crane part, aside from colour aliasing it also has some halo effect that is not present in the images. Sharpening was here set to zero, so it is not a sharpening artefact.

RawTherapee	CaptureOne 7.3
Lightroom 5.6	AccuRaw 2.2.6

The second crop is a more normal part of the image.

These conversions were not intended to be artful, the intention was to all raw developers near deafults. I tried to turn sharpening off in all developers.

Best regards
Erik

[Doug Peterson]

Two thoughts:
- Defaults are meaningless. How do the files handle under each processor when you're trying to get the best from each?
- Look at those smoooooth diagonal lines on the C1 version and see how none of the others come close to rendering those diagonal lines with such smoothness? This is the sort of thing that matters a LOT when you go to print big, especially if you need to interpolate upward to get there.

[synn]

LR is doing a horrible job with reds. Usual for it and one of the main reasons I am never using it again for anything serious.

The color artifacts in C1P can be fixed in a minute with an adjustment layer.

[eronald]

Two thoughts:
- Defaults are meaningless. How do the files handle under each processor when you're trying to get the best from each?
- Look at those smoooooth diagonal lines on the C1 version and see how none of the others come close to rendering those diagonal lines with such smoothness? This is the sort of thing that matters a LOT when you go to print big, especially if you need to interpolate upward to get there.

Doug,

 Yes, the diagonals in the C1 image have been "antialiased" by blurring.
 In fact, all these images look like they are from Blurristan, except the one from AccuRaw.
 Since I got my Sigma DP3, I suddenly started having images that look "naturally" sharp. Once you have Bayer, whatever you do there is a tradeoff.

Edmund

[eronald]

LR is doing a horrible job with reds. Usual for it and one of the main reasons I am never using it again for anything serious.

The color artifacts in C1P can be fixed in a minute with an adjustment layer.

You mean to say that software written by John Knoll and sold millions of times is not good enough for you when it is good enough for everybody else?

Edmund

[synn]

You mean to say that software written by John Knoll and sold millions of times is not good enough for you when it is good enough for everybody else?

Edmund

I mean the raw converter written by phase for phase cameras is doing a great job on phase files.
Anyone with more than 10 minutes of experience with medium format digital may feel free to correct me.

- fully paid up user of photoshop CC

[synn]

I feel silly trying to change the mindset of a keyboard warrior, but Edmund, feel free to pixel peep here:

http://flic.kr/p/oTkAsk

Tell me if you see any blurry lines.

[Ken R]

...Here we go again... Il bite.

Erik, thx for posting the images.

The C1 Pro conversion seems to be the most balanced. Good color, detail (without much artifacts) and texture.

I can see in the crops posted that LR is rendering the reds a bit more orange and with a bit more punch. Im sure that can be adjusted easily.

Overall I dont know if these images are the best for a comparison test of converters but they do show differences.

[ErikKaffehr]

Hi Doug,

The main idea with this is to look at the demosaicing in each raw processor. As a matter of fact, RawTherapee allows choice of several demosaicers. Defaults are not necessarily a bad choice as a starting point.

Doing the best of each raw processor requires practice. Also, these raw processors are intended for different purposes, LR is both a flexible raw processor and a DAM-tool, Capture One is similar, RawTherapee is a flexible platform including a set of demosaicers, AccuRaw uses high precision floating point algorithms trying to achieve best reproduction. RawTherapee and AccuRaw are intended to be used with an image editor like Photoshop, while LR is intended to do parametric editing, with some trips to Photoshop.

The reason I posted this is that I feel the demosaicer is a weak point of Lightroom and I may feel Adobe may need to do something to improve it. On the other hand the developers of RawTherapee and AccuRaw have some pride in their demosaicers.

My impression here is that LR fails on two issues, bad handling colour errors in the demosaicing. AccuRaw is essentially best on this.

I enclose two marked areas, of some interest. The green area shows the ribbed structure these buildings have. AccuRaw has here credible detail.  The roof, marked in red, shows some spatial aliasing on all images, neither developer can remove it. AccuRaw extracts more detail but there is a price in staircase artefacts.

This is my "official" processing of this image:



Best regards
Erik

Two thoughts:
- Defaults are meaningless. How do the files handle under each processor when you're trying to get the best from each?
- Look at those smoooooth diagonal lines on the C1 version and see how none of the others come close to rendering those diagonal lines with such smoothness? This is the sort of thing that matters a LOT when you go to print big, especially if you need to interpolate upward to get there.

[Jim Kasson]

Thanks, Erik. Have you tried Iridient Developer? Of course, that would give you not one more algorithm to try, but several. If you haven't tried it, you might want to wait for version 3.0 to get out of beta.

Jim

[eronald]

My impression here is that LR fails on two issues, bad handling colour errors in the demosaicing. AccuRaw is essentially best on this.


Best regards
Erik

I agree with you. Lightroom is not really a contender, C1 is ok, but AccuRaw seems to do best, at least on this image.
Frankly, I would have expected C1 to go the extra mile in sharpness on a Phase file, but since a few versions down the line my impression is it has become slightly more blurry. I ran a lot of P45+ files through it too, and for Phase files it consistently beat Lightroom in color and detail.

Edmund

[ErikKaffehr]

Hi Edmund,

I am not concerned as much about colour as about demosaic artefacts. I would suggest that excellent images are possible by John Knoll's software, but I would suggest the samples here indicate that there is room for improvements. The colours here are not red, BTW, the pilot boat is orange red and the houses are "falu red", closest colour is probably: NCS: S 5040-Y80R with Lab coordinates:

33.11
26.33
17.08

Several variants exist: http://falurodfarg.com/eng/product-information/colours/

Best regards
Erik

You mean to say that software written by John Knoll and sold millions of times is not good enough for you when it is good enough for everybody else?

Edmund

[ErikKaffehr]

Hi Jim,

I simply forgot about it!

The way it is, I am quite happy with Lightroom, except the issues illustrated here, both the colour artefacts but even more of the double contours. I guess that the demosaic algorithm is optimized for OLP-filtered images.

Now, what I see is that colour demosaic artefacts are quite common, and it seems that LR does a particularly bad job on those. To avoid those I would say a properly OLP filtered sensor with optimum pixel size is needed.

One of the crops shows what is essentially monochrome aliasing, one of the roofs.

I have tested Iridient developer several times, and I have some respect for it, but I don't feel it fits my needs. But definitively worth trying.

The way I see now, I will continue to use Lightroom and use it regularly, but if I have an image that I want to optimize for large print size I will probably go with some other tool for raw conversion.

One point is that I can use my DNG-profiles with RawTherapy, AccuRaw and Iridient RawDeveloper.

Best regards
Erik

Thanks, Erik. Have you tried Iridient Developer? Of course, that would give you not one more algorithm to try, but several. If you haven't tried it, you might want to wait for version 3.0 to get out of beta.

Jim

[ErikKaffehr]

Hi,

Just to say, the images here were enlarged to 200% using bicubic interpolation for easier viewing. Just looking at 200% in Photoshop uses nearest neighbour which will give some more artefacts. Clearly stated in my posting. Also the images are with sharpening set to zero.

One of the points here is that I want to have an artefact free conversion and sharpen using FocusMagic, when printing large (say 70x100 cm).

Best regards
Erik

I feel silly trying to change the mindset of a keyboard warrior, but Edmund, feel free to pixel peep here:

http://flic.kr/p/oTkAsk

Tell me if you see any blurry lines.

[ErikKaffehr]

Hi Ken,

I have chosen the image because it has a large amount of artefacts, and wanted to see if other converters handle it better.

The reds here are quite a bit orange (on the boats), it is intended for high visibility and used to be either greenish yellow or orange red.

Best regards
Erik

...Here we go again... Il bite.

Erik, thx for posting the images.

The C1 Pro conversion seems to be the most balanced. Good color, detail (without much artifacts) and texture.

I can see in the crops posted that LR is rendering the reds a bit more orange and with a bit more punch. Im sure that can be adjusted easily.

Overall I dont know if these images are the best for a comparison test of converters but they do show differences.

[Slobodan Blagojevic]

... the images here were enlarged to 200%...

I am confused... your screenshots say "100%"!?

[ErikKaffehr]

Hi Slobodan

Image->Size 200% (bicubic), intended for better viewing. It is a bit obscene, but I want to show what is going on and avoid the artefacts caused by zooming in.

Best regards
Erik

I am confused... your screenshots say "100%"!?

[Paul2660]

Erik:

From looking at the raw in C1 vr8, I find that C1 can do a good job even on the distant crane.  At first I did not realize how far back in the image the cranes were. 

I took the image up in C1 8, worked the sky and foreground in separate adjustment layers, then quickly ran both Topaz Clarity and then Focus Magic on the exported image from C1. 

Observations:

You have an excellent example of a P45+, I could not have shot that scene with mine and held both the clouds (no highlights blown) and still have as good shadow details on the buildings on the right and boat hull on the left.  Just would not have been able to get that with mine.  So pretty clear to me, that as Phase continued to develop the P45+, they made some significant under the cover improvements or I just had a pretty damn bad version. 

As for the Sigma, I don't agree at all.  Sure the file may seem sharper, but it's a lot smaller file to start with.  I personally don't feel any of the Sigma Foveon's could get to the size of the 'standard output" from a P45+ and hold up.  You would have to uprez the Sigma file considerable and I have yet to find any uprez solution that keeps the image looking as sharp as the original at 100%.  Sure the Foveon shot looks sharp, and more than likely will defeat anything in the same MP class in a 1.5 crop sensor at a 100% view, but the P45+ file holds a lot more room to work with, in regards to larger printing. 

Overall an extremely clean iso 50 image and I am impressed with the lens/camera combination as the cranes are a ways off and you still can make out the railing with no problems. 

I have attached a jpg of my attempt, using the steps I listed above.

Paul

[ErikKaffehr]

Hi Paul,

Thanks for the comments! Regarding exposure I enclose a screen dump from RawDigger.

I have used C1 v7.3, for this comparison. C1 v8 doesn't work on my version of Mac OS/X (10.8.5) AFAIK, and I don't really want to upgrade before I am forced to.

Best regards
Erik

Erik:

From looking at the raw in C1 vr8, I find that C1 can do a good job even on the distant crane.  At first I did not realize how far back in the image the cranes were.  

I took the image up in C1 8, worked the sky and foreground in separate adjustment layers, then quickly ran both Topaz Clarity and then Focus Magic on the exported image from C1.  

Observations:

You have an excellent example of a P45+, I could not have shot that scene with mine and held both the clouds (no highlights blown) and still have as good shadow details on the buildings on the right and boat hull on the left.  Just would not have been able to get that with mine.  So pretty clear to me, that as Phase continued to develop the P45+, they made some significant under the cover improvements or I just had a pretty damn bad version.  

As for the Sigma, I don't agree at all.  Sure the file may seem sharper, but it's a lot smaller file to start with.  I personally don't feel any of the Sigma Foveon's could get to the size of the 'standard output" from a P45+ and hold up.  You would have to uprez the Sigma file considerable and I have yet to find any uprez solution that keeps the image looking as sharp as the original at 100%.  Sure the Foveon shot looks sharp, and more than likely will defeat anything in the same MP class in a 1.5 crop sensor at a 100% view, but the P45+ file holds a lot more room to work with, in regards to larger printing.  

Overall an extremely clean iso 50 image and I am impressed with the lens/camera combination as the cranes are a ways off and you still can make out the railing with no problems.  

I have attached a jpg of my attempt, using the steps I listed above.

Paul

[synn]

Erik, let me have a go at the file later today. Will post the results up later.
I don't understand though, that you insist on keeping everything at default, but push sharpening to zero. All bayer files are assigned a default sharpening level by editing softwares upon import and for good reason. pushing them downwards to zero is not a reflection of any real world outcome.

If I were to do this test, I would edit all files to the best of what the softwares can achieve. doing four default renders is akin to driving a bunch of supercars down a city street to assess what their engines are capable of.