|
Forum: Digital Processing, Software, and Printing
03-15-2010, 03:51 AM
|
| |
(While it it ultimately unclear whether the following impressions relating to "raw" processing of Panasonic DMC-LX3 ".RW2" image files directly translates to how DxO behaves with Pentax "raw" image-files) ...
after gaining more "mileage" processing DMC-LX3 RW2 shots (taken at ISO Sensitivities up to 400), I am finding that (in most cases) I am preferring to "goose" the "Details" slider-control from the "auto" default up to 50 out of a total of 100 (as opposed to using the USM tool). A setting of 100 almost invariably is too much. This does usually result in a signal/noise ratio degradation due to the "deconvolution deblurring" (readily visible at 200% preview, and sometimes visible at 100% preview).
However, (from a person who in general has in the past abhorred any/all Noise Reduction algorithms in any/all "raw" processing), I am finding that judicious use of the NR tool (usually less than one-half of the amount that the DxO "Noise" tool automatically selects, and often decreasing the level of "Luminance" NR slider-control relative to the "Chrominance" NR slider-control somewhat) usually manages to deal fairly well with the increased image-noise resulting from the above-described "goosing" of the "deblurring".
I am preferring the use of the above regimen described above to the use of the DxO USM (which is performed on the full-sized image being processed) to increase a sense of "acuity" for images which are well-focused and relatively free from "camera-shake". When such is not the case, I do use the USM in addition to the "Lens Softness" correction - but leave the "Details" slider at it's default value of Zero (as it is in the less-focused areas where the signal/noise ratios of the "deblurring" appear to degrade the most).
It is very hard for my eyes to be able to tell whether DxO (in this case of ISO less than or equal to 400 using the DMC-LX3) is (or is not) applying some amount of Noise Reduction (when using the "Lens Softness" correction, and when the NR Tool is specifically de-activated). Thanks again to Falconeye for alerting me to this possible behavior! As it turns out, (due the signal/noise ratio exhibited by the DMC-LX3 image-sensor), it is unlikely that I will be venturing into ISO=800 territory with the DMC-LX3, anyway. But, (as my DMC-FZ50 is limited to ISO=200 for good results shooting in "raw" without having to resort to NR in post-processing), I am fairly happy that the DMC-LX3 shooting in "raw" appears to be fairly well-behaved up to ISO=400 (though I would not say the same for the in-camera JPGs).
On balance, however, (use of very moderate settings of) the DxO NR Tool (with the very welcome separate "Chrominance" and "Luminance" controls) is the first time that I have found myself able to tolerate the affects of a specific Noise Reduction algorithm! While NR is not something that I am finding to be necessary in all cases (when processing DMC-LX3 shots up to ISO=400), the net effect (of a bit of NR applied, often in order to correct for the artifacts from "goosing" the "deconvolution deblurring") appears (to me) to be a worthwhile trade-off.
Though I have not tried "Noise Ninja", I found the effects of "Neat Image Pro", as well as the effects of the "Silkypix 3.x", NR Tools to be too heavy-handed, and therefore unacceptable. (Thankfully), DxO NR (at least, when used in sparing amounts) appears to offer a better result (preservation of image-details) overall (with or without the above-described "goosing" of the "Details" slider-control in the "Lens Softness" tool).
As a further note: I generally re-sample the processed images (using Lanczos-3) to a 1600x1200 (or at times 1024x768) pixel-size, and apply mild USM at the final target-size (pixel-radius = 0.5, strength = 50% to 100%, threshold = 1% of full-scale) prior to conversion to JPG (using no chroma sub-sampling). The above impressions should be read with these additional factors in mind ... :)
|
|
Forum: Digital Processing, Software, and Printing
02-21-2010, 09:41 AM
|
| |
Falconeye,
Interesting. Thanks for your very clear and informative responses. You are correct about the four independent DOP6 crops - now I get it ... sorry ... :confused:
Some thoughts that might have some merit:
In the ""Lens Softness"" palette The "Global" slider (-2.0 to +2.0 range), is set at Zero, and the "Detail slider" (0 to 100 range) is set at Zero. Only then will the "Auto" annunciator display the AUTO" symbol.
I often try to adjust the "Detail" slider to a 50 position (out of 100), with 100% usually being artifact-laden "overkill". Better to use USM or a bit of that 'Local Contrast"), instead to enhance perceived "acuity".
If either of the above two controls modifies the "feedback-constant" of any (iterative) re-circulations of data (derived from any particular delay-tap position within the digital filtering chain that performs iterative de-convolutions of the incoming image data -
Then it (may) be that the adjusting these controls (to some extent, anyway) affects the "feedback-factor(s)", and in doing so, possibly the stability of such re-circulating (and thus compounding) errors arising out of the iterative process(es).
Note: The "Detail" slider only increases higher spatial-frequency information (and thus only would tend to worsen such artifacts, and it is only the "Global" slider that can be manually adjusted to negative values between 0 and -2.
I wonder if the above ("Global") slider decrease to values less than Zero might have an effect along the lines of the "missing control" - one that might have an ability to affect the "regularization of a deconvolution iteration (NR1)" as you describe it?
Please feel free to discount this (only-partially informed) hypothesis. I would be interested in what you think about whether downward adjustment of the "Global" Lens Softness Slider" reduction would serve to reduce the effective feedback-factor of such troublesome "recirculations". Of course, this might well appear (by some relationship between control setting and effect), to also reduce the magnitude of the ("global") corrections applied. Thus there really is no "free lunch"... ;)
Best regards,
DM |
|
Forum: Digital Processing, Software, and Printing
02-20-2010, 09:26 PM
|
| |
DM: Thanks, Falconeye. I am a bit confused, though. Your notations state that the ISO was in all cases 1600. But, there are two individual sets of two of DxO Optics Pro 6 images - and they appear different. I understand the differences relating to the DxO Settings described alongside the images. But there are two sets of them, and they appear different in the severity of their effects. Are those differences (perhaps) the result of different camera ISO Sensitivities? Or (perhaps) something to do with different levels of DxO adjustable settings? Please explain.
DM: Are you saying that what you originally thought was the effect of NR being active (even though disabled in the user interface of Version 5.x) appears to you now to (instead) have been the effects of "deconvolution deblurring" (DD)?
Or are you saying that you think that NR is still (in the current DxO OP 6) remaining active (even though disabled in the user interface), and that "artifacts" from "DD" are also contributing to the perceived visual "distortions"?
Thank you if you could clarify these things, and thanks again for your efforts and images posted!
|
|
Forum: Digital Processing, Software, and Printing
02-20-2010, 09:09 PM
|
| |
A few weeks ago I downloaded the trial version of FocusMagic 3.2 at:
http://www.focusmagic.com/download.htm
Installed it to try and fix pretty horribly out of focus shot that likely also had some camera-shake.
Can't say that I gave it much of a chance - it yielded no useful results at all with image I was trying to fix. But, the image was likely irretrievable by anything, anyway ... So that was not a fair trial.
I did note that it appeared that the release may have been a bit dated, has kind of a clunky little interface, and costs $45 USD. Having spent all my money on a DMC-LX3 and DxO Optics pro 6.x, I just uninstalled it and forgot about it.
Sounds like Falconeye has had some good successes using it, however. It's hard to argue with (anyone's) success ... :)
There is not much of anything else like it (a stand-alone program that can also act as a plugin for my beloved PaintShop Pro 9.011). Adobe (I read) has for some time now forced all their plugin marketers to agree to make their plugin incompatible with everything other than Adobe products. Thus, there is virtually nothing that can be used with my archaic PSP9.
I wonder if this might be the story behind FocusMagic not being updated for a while (if that is in fact the case). The installation file yields no clue as to it's original date of origin, whatsoever.
(Perhaps FocusMagic) preferred not to "play ball" with the jerks at Adobe and their exclusive legal agreements - and has continued to market their older version so as not to be controlled by Adobe?
|
|
Forum: Digital Processing, Software, and Printing
02-19-2010, 11:36 PM
|
| |
Falconeye,
I found a DxO web-page that confirms that DxO (in, at least the particular product described below) combines "deblurring" and "noise reduction" operations simultaneously (in this case, in a manner that is openly disclosed to the user):
"DxO Image Science and the ZEISS Camera-Eyepiece DC4 White Paper
... increasing image sharpness may increase noise visibility. Noise visibility is a complex function of light level, ISO speed setting, the uniformity of objects in the scene, and eye sensitivity.
In order to resolve the complex issue of compromising between sharpening and graininess reduction, DxO Labs has developed an innovative method to easily characterize noise within an image. This measure, called BxU, relates to the size of the spot induced by noise, and allows the simple adjustment of sharpening and graininess reduction filters.
DxO sharpness correction locally adapts the level of deblurring measured with BxU, to apply very strict noise control, keeping noise below the level of perceptibility.
DxO graininess filter increases the signal noise ratio locally by smoothing homogeneous parts of the image while preserving details in textured areas.
The combination of DxO sharpness and graininess reduction typically allows an average increase of image sharpness by 1.5 BxU and signal to noise ratio (SNR) up to 5dB (depending on ISO setting), without loosing details."
http://www.dxo.com/us/embedded_imaging/home/zeiss_dxo_camera_eyepiece/zeiss_camera_white_paper
DM: Thanks again for reporting your impressions that DxO Optics Pro 6.x (at least, at some unknown ISO Sensitivity, or, perhaps, at some detected threshold image noise level) may be silently applying Noise Reduction when the "Lens Softness" corrections are being performed - even if the "Noise Reduction" tools are disabled in the user interface.
I will be watching out for that - thanks again for your thoughts ... :)
|
|
Forum: Digital Processing, Software, and Printing
02-19-2010, 12:05 AM
|
| |
Falconeye,
Thanks for the Imatest information. Interesting, indeed!
Perhaps you know more than I about the details of the many simultaneous and inter-relating operations that DxO Optics Pro performs in implementing "Lens Softness" corrections. When you say:
" .. in whichever way you turn the available information, it is just short of being enough for a competitor to copy it. The required missing bit of information is the so-called regularization method applicable in iterative deconvolution."
I am not so sure that what is easily available to any plebeian such as myself via search engines is even coming close to summing up matters surrounding what (precisely, and in what manner) is going on within the inner-workings of DxO Optics Pro.
There is much that I do not know. From my background in electronic design and programming, I can tell you that it takes a great deal of design and development effort on the part of a number of people working simultaneously to craft, test, and fine tune even the simplest product implementations ... and there is nothing simple about this stuff ...
I would think that DxO would likely need to worry more about industrial espionage, defecting engineering employees, or someone managing to directly de-compile their object code contained in the program itself. I read (in a recent press release about their engineering work to develop the "camera-phone" technology - which appears to me to utilize an inherently different technology than the Optics Pro routines) that they are a fairly small company (with around 50 employees). That does not necessarily mean that they do not have the financial means to acquire and to subsequently enforce patent protections, however.
Additionally it is entirely possible that DxO is not the "only kid in town" who researches and applies "deconvolution" techniques to "de-blurring" operations. In the statements (allegedly) made (on 03-06-2006) by Tom Lianza, Director of Display and Capture Technologies, GretagMacbeth LLC, he also stated:
"Deconvolution routinely occurs in high end point and shoot cameras as well
as in some D-SLRs. The manufacturer has knowledge of the lens and detector
MTF response and they can perform an appropriate inverse filter. When
properly done, in-camera sharpening can be quite effective in removing the
effects of lens and sensor blurring."
http://www.velocityreviews.com/forums/t423088-deconvolution-software-any-practical-value.html
I personally like the folks at DxO and wish them well. They have treated me pretty well as a paying customer, and it is clear that they have some innovative and bright people working for/with them.
My personal feeling is that they (DxO) have little to fear from ordinary folks (like myself) posting what we find on the internet using search engines. If such is indeed the case, they are perhaps not quite as bright as they appear to be ... :)
|
|
Forum: Digital Processing, Software, and Printing
02-17-2010, 03:06 PM
|
| |
<The following quote: "Reducing the 'lens softness' or 'blur' or 'lack of sharpness' means performing local, color-channel dependent and anisotropic deconvolution of the image"
isn't anymore on DxO's web site.>
DM: The DxO page states that the document is being edited - so it may re-appear at some point with some interesting information.
<The only other reference on DxO's site is a research paper about how to make use of a large designed-in logitudinal aberration to improve fix focus cell phones>
DM: I saw that paper - but it describes something different than what we are discussing.
Here is the one statement on the internet that I can find that may have been the source for the comments that we have seen:
"DxO Labs has developed a unique unit called the BxU (Blur eXperience Unit) which is a mathematical way of describing this 'blur'. Reducing the ‘lens softness’ or 'blur' or 'lack of sharpness' means performing local, color-channel dependent and anisotropic deconvolution of the image produced by the camera."
http://www.beautiful-landscape.com/Thoughts35.html
DxO themselves provided me with this other link from the "beautiful landscape" web-site:
http://www.beautiful-landscape.com/Thoughts81-Jim%20Scott%20Panostitch.html
so they must know about his statements publicly made.
Here are two sources that include discussions about "anisotropic deconvolution" of images:
http://books.google.com/books?id=wp85FltDnbIC&pg=PA289&lpg=PA289&dq=%22aniso...age%22&f=false
http://www.cs.tut.fi/~foi/papers/SMMSP2004-IHT_Anis_LPA-ICI_Deconv.pdf
And a free set of MATLAB routines that demonstrate some of these processes:
http://www.cs.tut.fi/~lasip/2D/
A lot of this stuff is "way over my head" mathematically (and perhaps also a bit technical for others, as well). However, there are small bits of them from which I learn a little more about this interesting field of endeavor ...
.
|
|
Forum: Digital Processing, Software, and Printing
02-16-2010, 11:00 PM
|
| |
deejjjaaa,
I seem to be one of the unfortunate folks who has a PC (with WinXP Pro SP3) that crashes very frequently when trying to use RT 2.41.
Replacing "librtengine.dll" in Version 2.41 with the earlier developmental version of "librtengine.dll" (4,474 KB file-size) seemed to help (briefly).
However, after several clean re-installations of RT 2.41 (even when restored from previous backed-up boot-partitions, and with the older version of "librtengine.dll" copied over the the version installed in RT 2.41 before ever using the program, and with a re-boot immediately after copying-over "librtengine.dll"), the crashes continue, rendering RT 2.41 essentially unusable ... :(
I tried a clean install of the newer 3.0a1 effort - but it will not even run on my system (generates an error at program start-up that precludes it from even running).
I asked "dosdan" whether he knew of any further "magic DLLs", etc. that may be known to address. He does not know of any.
Saw one fellow (on the RT Forum) who stated that (perhaps some) of the Windows crashes had to do with multiple image-files existing in one folder directory. However, my problems have presented themselves with only a single, lone (DMC-FZ50 ".RAW") image-file (which is stated on the RT web-site to be a supported camera). I cannot find any other threads on the RT Forum that seem to suggest any other fixes or work-arounds for these crashes.
I was wondering if you (possibly) might know of any other measures (to try to get RT 2.41 to work) that I might possibly be able to try (as it appears that a stable, finished RT Version 3 is a long way from fruition) ...
DM
.
|
|
Forum: Digital Processing, Software, and Printing
02-16-2010, 10:37 PM
|
| |
DM: DxO does make public some information which may be the source of the statements that I (and it sounds you) have seen made by sources other than DxO.
The DxO information (quoted in bold blue italics below) is at: http://www.dxo.com/en/photo/dxo_optics_pro/optics_geometry_corrections/lens_softness
"DxO Optics Pro's fully automatic correction of image softness accounts for all the factors described below:
>any capture settings (aperture / focus distance / focal length, etc.)"
DM: Interestingly, while I recall reading that earlier versions of DxO Optics Pro (3.x, I think) actually required that the user input the focus distance (if the camera was not one that was able to note and record that data), Version 6.1.2 works with the DMC-LX3 (which definitely has no clue as to the focus-distance as a result of it's contrast-detection auto-focus, and thus could not be recording focus-distance data).
">any type of non-uniform softness (variation in degree and direction of softness)
>independent correction for all 3 color channels"
DM: Additionally, if one "Googles" the words [ "DxO" and "deconvolution" and "color channel" ], Google's sampled text from the DxO web-page content states:
"DxO Labs - Simply better images. ... means performing local, color-channel dependent and anisotropic deconvolution of the image produced by the camera. ..."
http://help.dxo.com/faq/index.php?action=artikel&cat=1610&id=16100074&artlang=en&highlight=lens
DM: The final statement from the text of the (first) DxO URL above states:
">very strict control to keep noise below perceptible levels"
DM: Additionally, from the text of the (first) DxO URL above:
"5. Noise dependency
... an image sharpening algorithm needs to adapt the level of deblurring locally so as to keep noise below the level of perceptibility."
DM: Thus, your operational observations (quoted below) seems alluded to in the DxO provided information (quoted above):
DM: Regarding Photoshop's "smart sharpening" (in the year 2006, anyway):
DM: I found this statement (allegedly) made (on 03-06-2006) by Tom Lianza, Director of Display and Capture Technologies, GretagMacbeth LLC:
"The sharpening algorithms used in photoshop (with the exception of unsharp
masking) are a form of deconvolution, but with an arbitrary estimate of the
blurring function. Unsharp masking is an image dependant operation, while
the deconvolution operations are independent of image data."
http://www.velocityreviews.com/forums/t423088-deconvolution-software-any-practical-value.html
.
|
|
Forum: Digital Processing, Software, and Printing
02-16-2010, 01:55 AM
|
| |
falconeye,
I have recently purchased DxO Optics Pro 6, and am now using it with my Panasonic DMC-LX3 RW2 "raw" images. Have processed a fair number at this point, and I am liking what I see ...
I have taken great interest in DxO's implementation of "deconvolution deblurring" in their Lens Softness" correction (for module-supported "raw" only).
You seem to know more about the specifics of this than many folks. The DxO web-site says little or nothing. A DxO Tech Support Rep spoke a bit about it a replying email. I recently found this statement:
.... " ... that fix for softness is really just deconvolution. However, unique to DxO, their deconvolution solutions are calibrated for each color channel (RGB), calibrated across the focal length of the lens, and calibrated across the entire field of view." ...
(-b shaw, Sep 4 2009, 02:56)
Adobe CR v DxO - Dyxum forums - Page 1
Do you know whether this is accurate information quoted above? Do you know any other interesting facts about "DD" in DxO's "Lens Softness" corrections?
Also, you state (above in this thread):
"DxO then applies aggressive noise reduction at higher ISO levels (deconvolution would mathematically become unstable otherwise) and it does so even if noise reduction is disabled!"
How do you know this? From solid personal processing experience? From another source? I have assumed that the NR unchecked means "NR Off" ... If this is the case, do you know (or have you observed) any particular ISO levels (and for which camera/lenses, I suppose I should also ask) at which this "silent NR" begins to take place?
You might find some of my posts interesting (with sample processed RW2 "raw" image-files) about the DMC-LX3, Dxo Optics Pro 6 - and a few of comparison with the Silkypix DS 3.021 SE that comes with Panasonic ("raw" recording) cameras - on the thread currently ending at:
Ricoh Forum • View topic - Panasonic Lumix DMC-LX3 |
