[brntoki]

If you deblur an image such as the K-5 IIs, you end up with an overcooked image. And by overcooked, I don't mean razor sharp, but rather... an image that simply looks bad. The same can be said for an oversharpened image as well. ie. we can only amplify edge contrast or detail to a certain level before things begin to fall apart, and so, there are physical limits to what can and can't be accomplished this way.

That being said, I'm not sure a maximum sharpness presentation is the best approach to take in cases such as these either. Which is why I chose to limit my adjustments to removing the effects of the AA filter exclusively. This way, interested parties can take both images and process them accordingly to see how each holds-up. Otherwise... its unlikely that anyone will reach a satisfactory conclusions through the eyes of another person.

With this in mind, I'd say that the limitations associated with detail extrapolation give us the potential to determine just how much detail each image contains and more importantly... where detail cannot be created where none existed.

Hope this helps.

PS. these results are not exclusive to the K-5 II/K-5 IIs btw. As it turns-out the D800E is also outclassed by the D800 using the very same approach.

John Bee and Class A,

At first I admit I didn't get the idea of applying detail enhancing post processing to only the 5II. It didn't register with me at all. I do at least now understand your reasoning, and even admit it is interesting. So, I am more enlightened, I think, than before I came across these tests and ways of thinking, which is always awesome. So, thank you both.

Bottom line, yes, whatever post processing is necessary to either or both II and IIs images regarding detail/sharpness/acuteness/microcontrast/etc., it can be overdone. There will be a limit in both cases. I suppose the real test is how much apparent detail can be extracted from each with the best possible post processing applied.

I look forward to reading and considering more thoughts on this from either of you and others as well.

[kataria0]

The Bayer-AA-filter in front of the sensor blurs the image slightly in order to avoid colour artefacts. It is valid and recommend to undo that blur by a bit of capture sharpening (deconvolution sharpening would be ideal, but regular sharpening methods work as well).

A K-5 IIs image, on the other hand, did not receive a pre-blur that needs to be undone.

Class A, yes that is exactly the throught process behind my questions. Thank you for putting the details.

To everyone: With due respect I am interested in just applying a little bit of sharpness (not extreme) just enough to be pleasing to the eye on K-5 ii and compare with unsharpened K-5 iis.

Furthermore I am not looking for comparison on pictures of high contrast dense patterns as K-5 ii should naturally likely outperform the k-5 iis under such condition due to anti-aliasing filter.

One day I plan to use a very high zoom to take pictures of animals, sometimes they have dense pattern in their fur/feathers and am concerned about aliasing effects showing up otherwise I would go for K-5 iis. Also there are lots of palm trees where I live, and wonder if aliasing effect will show in the pictures of palm fronds using K-5 iis especially near the region where the frond pattern is dense near the trunk of the tree.
Such are my concerns.

[Class A]

@kataria0: Are you following the "real-life-moire-example" thread? John Bee has just posted another comparison that you may find interesting.

For me the question of whether or not to omit an AA-filter has become a non-issue.

1. The "out-of-the-box" advantage of a K-5 IIs is mostly relevant for pixel-peeping whereas real world moiré can be visible at standard magnifications.
2. Extra-capture sharpening can bring the K-5 II to K-5 IIs levels most of the time. I haven't checked whether John Bee's investigations are bullet-proof (his image source is not optimal for this purpose) but what I see aligns with what one would expect theoretically.
3. Fixing moiré manually would be too cumbersome for me and the results Adam presented look bad to me (in particular for people who pixel peep to identify the last iota of pixel-sharpness).
4. There may be a number of occasions where a K-5 IIs will indeed capture more detail, but I don't expect the combination of this happening with the actual need for that extra iota of detail to occur frequently.

Before we reach 100MP, or a departure from Bayer-CFAs becomes mainstream, my cameras will have Bayer-AA-filters. Other people's mileage varies, I'm sure.

[RBellavance]

For me the question of whether or not to omit an AA-filter has become a non-issue.
...
Before we reach 100MP, or a departure from Bayer-CFAs becomes mainstream, my cameras will have Bayer-AA-filters. Other people's mileage varies, I'm sure.

I have come to the same conclusion as you. If I go for a K-5 II (and I'm *really* tempted - except a part of me is yelling "wait to see what they'll announce next"), it will be the standard model, not the "s".

[kataria0]

@kataria0: Are you following the "real-life-moire-example" thread? John Bee has just posted another comparison that you may find interesting.

For me the question of whether or not to omit an AA-filter has become a non-issue.

1. The "out-of-the-box" advantage of a K-5 IIs is mostly relevant for pixel-peeping whereas real world moiré can be visible at standard magnifications.
2. Extra-capture sharpening can bring the K-5 II to K-5 IIs levels most of the time. I haven't checked whether John Bee's investigations are bullet-proof (his image source is not optimal for this purpose) but what I see aligns with what one would expect theoretically.
3. Fixing moiré manually would be too cumbersome for me and the results Adam presented look bad to me (in particular for people who pixel peep to identify the last iota of pixel-sharpness).
4. There may be a number of occasions where a K-5 IIs will indeed capture more detail, but I don't expect the combination of this happening with the actual need for that extra iota of detail to occur frequently.

Before we reach 100MP, or a departure from Bayer-CFAs becomes mainstream, my cameras will have Bayer-AA-filters. Other people's mileage varies, I'm sure.

Class A. Yes I just looked at the pictures. K-5ii picture was perhaps slightly out of focus or maybe it needed some sharpening. In any case I don't want to deal with moire either so K-5 ii is a good option for me and its en route for delivery.

BTW I believe you mentioned a software you use for deconvoluted sharpness. I have few questions
1. What is this free software name?
2. Is it stable? I have 64 bit XP and 4GB RAM on my laptop
3. When you present your raw image to this software does it pick up all the camera settings, focal length, f stop etc automatically from the header file?
4. It appears Photoshop offers such a sharpnees module as well. Would that be perhaps easier to run?

Thanks

[JohnBee]

Class A. Yes I just looked at the pictures. K-5ii picture was perhaps slightly out of focus or maybe it needed some sharpening. In any case I don't want to deal with moire either so K-5 ii is a good option for me and its en route for delivery.

BTW I believe you mentioned a software you use for deconvoluted sharpness. I have few questions
1. What is this free software name?
2. Is it stable? I have 64 bit XP and 4GB RAM on my laptop
3. When you present your raw image to this software does it pick up all the camera settings, focal length, f stop etc automatically from the header file?
4. It appears Photoshop offers such a sharpnees module as well. Would that be perhaps easier to run?

Thanks

Raw Therapee is a free Raw developer with a deconvolution sharpening tool.
It also runs on XP, though it is known to crash(or shut down) from time to time.
That being said, any sharpening tool would help, though a deblur tool works best for removing the effects of an AA filter.

[Rondec]

I think the basic question is whether the K IIs actually really captures any more detail than the K5 II does. If it doesn't, then the lack of an AA filter is only detrimental in the end. If it does, then many people will suffer the possibility of moire for the extra detail. From what I have seen, there certainly isn't a dramatic difference, even without deblurring.

[Michaelina2]

...a deblur tool works best for removing the effects of an AA filter.

Thanks for reminding me about Topaz DETAIL Got a copy of their recent update, DETAIL 3 (free to current owners ), and played around with it (via LR4.2) on some RAW Pentax K-5/Q files. Perhaps my skill level has changed so I can better appreciate what it does, or they have vastly improved the program. Anyway, it offers a very powerful alternative to traditional sharpening and I plan to use it more. Again, thanks for the tip.

Cheers...M

[brntoki]

JohnBee or Class A,

How does the deblur function actually work? You keep saying that it can put the 5II on an equal footing with the 5IIs, implying that, from there, it can be further tweaked and sharpened just as much as the IIs. But that seems hard to believe because as far as I can tell, deblur is also a manipulation of contrast to "hide" softness. Or, is that not the case? Does a deblur or deconvolution work in a totally different manner? And, if it does not work in a different manner, why cannot the 5IIs also benefit from its use? Or, why is it not made clear that it, too, will move a file closer to its limit of post process sharpening due to artifacting?

I would test this myself right now, but I don't have access to my graphics machine at the moment. Some pushing of the deconvolute sharpening in RawTherapee seemed to me to bring out "normal" sharpening artifacts as would other methods.

[Michaelina2]

... How does the deblur function actually work?

May I suggest that you register for one of Topaz's excellent (and FREE) real-time interactive demonstrations of DETAIL v3? You can message the instructor during the presentation and get an instant answer to your questions.

See...

Topaz Labs - Free Webinars and Live Training=

Cheers...M

[Class A]

How does the deblur function actually work?

Deblurring on the basis of deconvolution attempts to undo the effect of a known blur, e.g., that of a lens defocus. Let's stick to this latter example as it is more straightforward to discuss than the AA-Bayer-Filter case.

Imagine the recorded blurred image to be a combination of a clean signal (the image data as it would be recorded if every sensel could record all colours and no blurring would be needed) and some blur (e.g., caused by lens defocus). Mathematically, this combination is called a convolution. The point spread function of the blur is multiplied with every single "clean image" data point and the sum of all these blurred points is the resulting blurred image. You can reverse such a convolution with a deconvolution.

BTW, the same principle can be used to undo the effect of motion blur.

As such, deconvolution is fundamentally different (and computationally more intensive) than standard micro-contrast enhancement.

P.S.: Kudos to your attitude, i.e., being open to learning new things!

[brntoki]

Michaelina2, thanks for your suggestion. I'm not sure if I will take advantage of that opportunity or not, though, as I've just got a lot going on. Very good idea, however.

Class A, thanks. I think I understand what you're saying there and it seems to make sense. I'm not sure if I am warranted, however, to think of deconvolution sharpening as a contrast enhancement that is just smarter, with rhyme and reason, as it were. Computationally, at least theoretically, it can figure out what needs undoing, but when it goes about that undoing, is it accomplished with contrast adjustments or some sort of pixel shifting, or something else?

What I'm getting at, if it isn't clear, is that if it comes down to a contrast adjustment, whether smart or blind (and there is an advantage to being done smarter, for sure) of some sort, then fundamentally it is the same sort of adjustment as other sharpening methods, to include the same limitations.

Anyway, now I'll read the link you provided and probably have my questions answered. I suppose I did that a little backwards, but old habits are hard to break.

[brntoki]

OK. Interesting info on Wikipedia, but it doesn't say what or how an image is manipulated after these computations are made. I am assuming it is primarily (or entirely) a contrast adjustment. Sorry! I still don't have access to my editing machine.

[JohnBee]

OK. Interesting info on Wikipedia, but it doesn't say what or how an image is manipulated after these computations are made. I am assuming it is primarily (or entirely) a contrast adjustment. Sorry! I still don't have access to my editing machine.

Deconvolution sharpening ends with pixel shifting.


(1) unblurred point (2) simple lens blur (3) vertical motion blur (4) complicated blur/motion, lens imperfection etc

Basically deblurring is a reversal of blurring, which is why they are also refered to as restoration algorithms. Blurring is the result of spreading pixels into neighboring areas(think diffusion). A deconvolution process is about moving or returning pixels back to a given location. ie. a blurring component is mathematically modeled as a convolution, hence the term deconvolution for the inverse process.

PS. one of the challenges in such processes often comes-in terms of detail & noise. One preferrably sharpens the signal, not the noise. Which I've found to be an issue with tools such as Photoshop's smart-sharpen. - which is also based on a deconvolution process btw.

[brntoki]

Interesting, John. Thanks again.

I can now use my computer again and hopefully can play with RawTherapee's deconvolution sharpening a little more.

BTW, have you tried to take each of the 5II and 5IIs files and sharpen them as far as they both will allow and compare them? Also, something that may be interesting other than the problem of deconvolution also enhancing noise, is how it affects desired blur due to shallow depth of field. I suppose it shouldn't be a problem, but it would be interesting to see if there is any disadvantage there.