[Click here for a larger version]
[Click here for a larger color version]

This is a five image panorama of SR520 taken yesterday late afternoon (it’s dark in Seattle late afternoon these days) from the Washington Park Arboretum.

I stitched this together using Hugin which produced (the way I had it configured) a series of images ready to layer in to the final panorama. Normally Hugin does a really good job blending different exposures but I was in a bit of a rush (rain) so I forgot to equalize the exposures and therefore had to make some manual adjustments.

In a month or so I’m going to redo this one because I’m more bothered by the remnants of the water lilles scattered across the foreground than I thought I would be!

This is part 2 of a comparison between Raw Therapee (RT) and Adobe Camera Raw (ACR) although I’m not sure how much I will be able to call it a “comparison” by the time I’m done.

The image I used in the first part of this post was a seascape taken at night under moonlight. The tones were relatively even, the detail in the rocks was subtle, and there were no extremes of lighting and contrast to deal with. Nearly all of those things exist in the city so for this post I’m using an image from Amsterdam taken in March at around 4:30 in the morning while I was suffering from extreme jet lag and couldn’t sleep.

I mentioned that RT offers the choice between three different demosaicing algorithms. I converted the image in the prior post using HPHD and was happy with the results.

I had originally planned to show some of the shortcomings of HPHD in this new image but after zooming in and out, in and out, in and out, for way too long I think that unless you’re viewing the image at 200%+ the artifacts HPHD produces are not visible and there is no chance they would show up in a print. But for what it’s worth I’ll will post some of the nitty gritty details anyway - I think it’s interesting to see the differences even if they’re largely irrelevant in most cases.

So first here are some crops from the sky showing the two RT algorithms and ACR - all with an additional curves layer to enhance contrast.

The 100% crops show minimal differences between the RT algorithms but the noise reduction applied by ACR is obvious. At 300% the difference between the RT algorithms in a smooth area are visible. EAHD produces a cross hatch pattern while HPHD seems to be more natual looking and does not show the pattern. I suspect this difference would be more exaggerated in an image with more noise to start with. This difference seems to be similar to the crops shown in this thread between the DCRAW AHD algorithm and the PerfectRaw AFD algorithm which looks very promising.

If there were zero noise reduction in the ACR crop it would look similar to HPHD.

Here are crops of some of the areas that HPHD produced artifacts:

Like I said this is really picky and although they’re not visible at 100% they are there in the HPHD conversion and not in the others.

Here are crops of all three at 100% and they are very similar - there is perhaps slightly more visible detail in the bricks and ornamentation in the RT crops but also more noise.

After I posted a link to part 1 of this post on the Luminous Landscape I got some not totally unexpected feedback from a supporter of one of the products suggesting that it was a flawed comparison because I wasn’t comparing “the best” output from each tool (including sharpening, noise reduction, etc) and therefore the results had no relevance to the real world of processing images. I suggested that in many cases, “the best” (either in overall image quality or what most closely matches what I want from the image) isn’t something that I generally get either entirely in the RAW converter without additional processing nor is it normally the result of a single RAW conversion. Secondly if you’re only applying capture level sharpening to the image in the RAW converter isn’t it even more important to be able to turn off default noise reduction if necessary to retain all potential detail in the image prior to beginning any post processing? In general, “the best” I could get out of either of the tools is probably far beyond what I would want to use going in to post processing.

In any case, before I post some crops of the image where I think I got “the best” from each application I want to mention a couple things that are important.

1. My computer had smoke coming out of it before I was done (er, gave up) so if you’re going to even attempt to compare two applications each with the potential for hundreds of possible output variations (taking in to consideration the variety of noise reduction and sharpening algorithms, etc) be prepared. Also be prepared that according to someone out there you will have done it completely wrong.

2. The Richardson-Lucy Deconvolution sharpening option in RT is a nice addition because it behaves differently than either the USM sharpening in RT or the sharpening in ACR. I was able to get more detail out of the window ornamentation, the bricks in the wall, the building ornamentation, etc, than I was using the USM option in RT or the ACR sharpening. I believe ‘Smart Sharpen’ in Photoshop uses something similar but I was not able to match the results using Smart Sharpen in post processing (not saying it can’t be done only that I couldn’t do it). The RL sharpening fared best when I let it do it’s thing and then masked out the smooth areas afterwards in Photoshop.

So without further delay here is “the best” I was able to do with each of these converters. As usual, a gradient map has been applied on top to convert to black and white and for consistency with my other non-scientific tests. For both I used the same mask to mask out the sky from the sharpened result - neither tool provides enough control for “damping” or “masking” the application of sharpening so I chose to do that in Photoshop instead.

At 100% there are some clear differences and I think they are entirely attributable to the RL Deconvolution sharpening method available in RT. To me the details in the ornamentation and small areas where leaves/branches are visible look more natural and less like I tried to add sharpening to bring them out. In addition there is more detail in the bricks in the RT version.

Finally, despite all the work getting “the best” out of each converter, when you view the image at 50% or print size there is literally no difference at all

The primary gripe I’ve had with ACR is that it is impossible to turn off all noise reduction - a default level of noise reduction is applied even when the noise reduction sliders are set to 0. One would think that an advanced RAW converter would let advanced users take control of the whole process but apparently Adobe thinks they always know better and have therefore decided not to provide that option. To their credit, however, after the high numbers of complaints about ACR 4.1, since ACR 4.2 they have reduced the default amount of noise reduction applied in particular to high ISO images. I have not tried Lightroom 2.x but I assume it is the same (so someone please tell me if I’m wrong.)

Night photography presents special challenges to digital photographers who must carefully balance noise and detail through a combination of tools and techniques such as exposing to the right, averaging exposures, using post processing noise reduction (Noise Ninja, Neat Image, and others), etc. Having full control over the process is critical when seeking the highest quality in the final result - online or in a print. In some cases the default noise reduction applied by ACR can smear important details or generally produce a very “digital” looking image and whatever issues exist after the RAW conversion are amplified in post processing.

Despite that major drawback, in general I have found that for most images the output quality from ACR is very high and it certainly fits well in to my [for now] all-Adobe workflow. All other things being roughly equal I have to stick with the tools that make my life the easiest. Still, every once in a while I decide to try to find something better…

Raw Therapee (RT) is a free, fully featured RAW converter which uses DCRAW for decoding (like many other RAW processing applications) but has implemented their own demosaicing and image processing algorithms. I’m not going to write about features like file browsing, highlight recovery, and sharpening methods because I’m mostly interested in the quality of the output before all those other things are applied (with the exception of Adobe of course) which hopefully shows how the particular demosaicing algorithms handle each image. RT provides the option to use three different demosaicing algorithms - VNG-4, HPHD, and EAHD. HPHD is the default and at first glance seemed to be the best so I used it for this test.

The image I’m going to use is one of the three images, each ISO 400, that were averaged together to create the final image in this post. One important thing to mention is that my current camera (hurry up with that 5D Mark II Canon!), the Canon Rebel XT (350D) is a couple generations old and the “quality” of the noise has improved considerably in newer generations and surely the converters are able to produce higher quality output with higher quality input…or at least one would hope!

Processing parameters were matched as closely as possible - all the exposure and histogram related settings were set to defaults in ACR and I did my best to match them in RT where only minimal changes were required and certainly nothing that would meaningfully affect output quality. Luminance noise reduction and sharpening were turned off as completely as each tool allows (remember my primary gripe with ACR.) If there’s any question as to how the settings were configured, here are some screenshots. You can see the only thing I needed to do was to adjust the luminance curve in RT slightly to match the default ACR conversion.

Raw Therapee:

ACR (corrected - the previous image showed sharpening of 2 which is incorrect):

The ACR image was opened directly from ACR in to Photoshop and the RT image was saved as a 16 bit TIF and then opened in Photoshop. I layered the images on top of each other and adjusted slightly to make sure they were correctly aligned.

Here are crops from the horizon from both RT and ACR. I added a gradient map layer to convert to black and white (the method I most commonly use these days) and curves layer to increase the contrast to show the differences more clearly.

A couple of things are significant - the first is the hot pixel in the RT crop which ACR removes quickly by default and RT does’t attempt to remove. My camera produces quite a few hot pixels in a long exposure so removing them manually would be a hassle. The second is the default noise reduction that ACR applies. If you look closely you can see some of that “digital” look showing through. As I’ve said, this is something I would rather avoid from the outset by being able to turn the noise reduction all the way off. No sharpening has been applied to the image at this point but there aren’t really any details here to sharpen so the only result would be additional emphasis on the noise in the case of the RT crop and more artifacts and digital look for the ACR crop.

And just for reference, here it is sharpened (USM amount 300, radius 0.7) and as you can see there is a reason it’s generally best to constrain sharpening to areas of an image that would benefit from it.

This is a crop from the rocky area. The same gradient map was used to convert the image to black and white but a different curves layer was used to lighten area of detail.

Again the default noise reduction applied by ACR is very obvious (keep in mind this is a 100% crop) as well as the appearance of more detail in the RT crop. This is a perfect example of a case where I would like to be able to disable noise reduction or even better be able to apply it selectively in the RAW converter. The next crop is the sharpened detail in the rocks.

There is no doubt that the RT crop appears to show more detail than the ACR crop - and as stated previously I believe that is a result of the default noise reduction in ACR. Again, this is a 100% crop in Photoshop so any print would not show this level of difference however I still think it’s significant.

So as you can see for this particular image Raw Therapee works very well and in my opinion provides better results overall than ACR as long as the sharpening is applied carefully and selectively. This is not the case for all images though and soon (this took much longer than I originally thought) I will post another in-city image for which ACR seems to excel while RT struggles…stay tuned.

Brian

But just in case you (or any of your friends, pass this on) needed any further encouragement to vote down an amendment writing discrimination against a particular group of people in to the state constitution, please take a look at the craziness that is providing so much of the support for this proposition.

You probably recognize this location from either the previous blog post or possibly some other images on my flickr site. This was taken about five minutes before the image in the previous post. I was ready to head back to bed (three hours sleep isn’t quite enough) but turned around to see the moon beginning to shine though the clouds.

What I hoped was that the initial appearance of the moon was signaling a parting of the thick cloud cover and that in its wake would be an evenly dispersed layer of fast moving altocumulus clouds (yeah, right!) the results of which, when backlit by the moon, are seen in this image from Gasworks Park in Seattle in 2006. However, that night at Gasworks Park in 2006 was and still is the only time I have been so lucky.

I’m not complaining, however, because during the minute or so that the moon did appear I captured the only image I have captured thus far in which the moon looks like a comet shooting across the sky (thanks to those that pointed that out!)

Yachats, Oregon, is about a six hour drive from Seattle and if you get full cooperation from the clouds, the tides, the moon, etc, it’s worth every minute of it. But as rare as the confluence of naturally occurring factors is, it’s even more rare to see me up at five in the morning - I like to do the night photography thing before going to bed.

The sun was set to rise at 7:30AM and the moon would set at 10:30AM so the moon would still be high in the sky around 5:00AM and would be bright enough to back-light the clouds. The clouds turned out to be a little thicker than I hoped but there was still enough light to provide backing to the clouds and also to squeeze in a couple night photography hours before the sun rose.

This is three 90 second exposures at ISO 400 averaged to reduce noise. Shorter, averaged exposures are often a better option on the coast because you never know (if you’re close enough) when you’ll have to grab everything and run to get out of the way of the ocean. It’s better in my opinion to use a higher ISO and stack/average n well exposed images to improve the quality of a “good” exposure than to try one really long exposure at the lowest ISO that only lasts half as long as it needs to. A single image at ISO 400 using the “expose to the right” technique will be better than an under-exposed ISO 100 exposure pushed two stops to match.

I watched the testimony by Secretary Paulson and Chairman Bernanke on CSPAN today and while the questions and comments by members of the Senate Banking Committee were surprisingly tough (whether there actions follow their tone is still in doubt) the answers given clearly revealed exactly the kind of concerns many are been worried about…

• purchasing assets at prices much higher than their current market value (”fire sale prices”)
• no executive compensation limits
• no penalties for those institutions that accept the help
• support for foreign financial institutions operating in the US (where are their Treasury departments?)
• opposition to needs based incremental implementation of the bailout
• opposition to restrictions on asset manager selection for managing the purchased assets (hmn, I wonder who will get the no-bid contracts)

…and on and on.

Also not surprising is the admission that the Administration has been sitting on this plan for weeks or months and now presenting it as an emergency plan to be implemented in the moments before a complete financial collapse…and then, after the bill is passed it will be time to debate it (from his testimony today).

A $700 billion (at its absolute minimum) bailout by the US Treasury, oops, I mean the taxpayer, of the “free market” with no checks and balances!??!?!? Handing over dictatorial control of our financial system to an ex Wall Street executive and explicitly declaring that no court of law or regulatory agency will have oversight? You’ve got to be kidding me!

This sounds suspiciously like the emergency implementation of the Patriot Act…

Step1: Whip everyone in to a panic frenzy…

Then: America is under attack, we must stop them before there is a mushroom cloud hanging over one of our cities (and it could be yours)!

Now: Our entire financial system is days away from complete collapse, there is no time for questions, you could lose your savings!

Real quotes from today: “The times are too urgent.” “…clean and quick [passage of the bill]”

Step 2: Urgently pass a vaguely written bill giving sweeping powers to the administration (the same administration I might add, not that this is a partisan post, it’s not) to “fix” the problems.

I suspect we’re all familiar with what happens later…

UPDATE 09/20/2008: There are signs of life in the Congress, hopefully they amount to something but I’m not getting my hopes up.

There have been few beneficiaries of the depressing “Spring” weather (the witty local press is calling it June-uary) in Seattle this year but the Calla lillies at the local p-patch (community garden) are on the list. Last year by this time they were completely destroyed by a variety of bugs that ate them to pieces before they could even open. This year they’re still nearly perfect and as usual they’re great subjects for photography.

A larger version of the first image is on my flickr site.

On occasion a type of noise appears in some digital images that is nearly impossible to remove using an automated process - in fact using “normal” noise reduction actually makes this type of noise stand out even more. I can’t really reproduce it because I’m not really sure exactly what causes it to appear. However, I know it exists and that seems to be particularly bothersome during long exposures in warm weather and that it shows up mostly in the mid-tones and shadows.

I’m certain it can be fixed at exposure time by one or a combination of ETTR, Zero Noise, or HDR techniques, however none of those are particularly suited for long exposures with star trails which is the example I’m using here. Perhaps image averaging would work but I wasn’t familiar enough with it at the time of this exposure to use it.

In any case, here is the original image I’m working with:

While you can’t see the noise in the image at this size it is very obvious at a normal print size and would ruin the image if it was not removed.

Here is a 100% crop of the original in the lower area of the rocks - you can see the white speckly noise very clearly:

And here is the same 100% crop with the noise removed - quite a difference:

So the technique to remove this noise is one of those that starts with “Well, I’ll just try it on a small area first…” and then “That looks great, and it didn’t take that long!” followed by “Wow, this is taking forever, I can’t believe I’m still doing this…” and finally “I’ve come too far to stop now…”

But at least it’s really simple - zoom to 400% in your editor and use the clone tool or the healing brush to click on every single little speck. Sound tedious? Well, it is. And it also takes a really long time!

All is not lost however…the newest generation Canons - probably those that are using the Digic III processor and newer sensors - seem to have a much better handle on at least one type of noise that is difficult to remove during post processing. Post #4 in this thread compares ISO 1600 images of the 400D and the 450D and it’s obvious that the 450D has a much cleaner image including better control of the type of noise I’ve been complaining about.

Obviously this is not for everyone and certainly not for more than one or two images a year but when you need a solution it does really work.