Resolution of M43 lenses

Started Mar 31, 2013 | Discussions thread
Detail Man
Forum ProPosts: 15,000
Like?
Re: Demonstration of Differences in DxO Geometric Corrections
In reply to Anders W, Apr 2, 2013

Anders W wrote:

Detail Man wrote:

Anders W wrote:

Detail Man wrote:

Anders W wrote:

Dr_Jon wrote:

I'd need to shoot something a bit more diagnostically useful. I can certainly say they aren't using the same distortion correction as the JPEGs. There is a free trial of DXO so others can play too...

I know DxO profile lenses themselves, e.g., for sharpening. But it would surprise me if what they do with regard to geometric distortion correction of RAWs from MFT equipment differs more than marginally, from what the in-camera jpeg engine or other RAW converters, e.g., LR would do. According to Lenstip, the in-camera jpeg engine leaves only -1.39 percent worth of barrel distortion from the 12-35/2.8 at 12 mm, which, as a rule, is barely visible. Your example sounded quite dramatic so could you please post the samples you were talking about and we'll have a look.

It could be that DxO tries to rescue more of the uncorrected frame than the jpeg engine or other RAW converters would. As you can see from the thread I linked to, it seems that the standard procedure is to chop off slightly more than would be required to get rid of the distortion. But I wouldn't think the distortion left behind by DxO is much different from that left behind by other processing alternatives for the simple reason that the latter get fairly close to the zero-distortion level.

I can confirm that DxO Optics Pro geometric distortion corrections (in the case of my LX3 and GH2 RW2s) consistently have a visibly notably different appearance than the Panasonic OOC JPGs as well as the nature of the geometric corrections that can be performed using Lightroom or Silkypix.

After recently spending a lot of time when posting on the referenced thread trying to get RAW Therapee's (manual) geometric distortion correction to look like DxO's, I have concluded that DxO's geometric distortion corrections are based upon more than geometric corrections performed relative to a single point in the center of the image-frame (as Panasonic's image-file meta-data as well as RAW Therapee's appear to be) ... and are instead based upon geometric corrections performed relative to multiple points existing within the image-frame (how many I do not know).

It may be that a metric of residual geometric distortion based on evaluation referenced to a single reference-point existing at the center of an image-frame may not serve to adequately describe residual geometric distortion that is referenced to multiple points within an image-frame.

It may well be that DxO uses a different method. But aside from rescuing a greater portion of the uncorrected frame than the OOC jpeg (which RT does as well), is there a noticeable improvement in the quality of the correction? That's kind of hard to judge based on the examples you posted.

The term "quality" may not be well addressed by numerical metrics here. It probably comes down to what various eyes see and minds perceive. That does not render the matter as "moot" - but it does involve subjective perceptions. To my eyes, DxO corrections always make the OOC JPG corrections (or single reference-point corrections performed by other RAW processors) looking bulbous and bloated at the center. Have not once found that I preferred such corrections to what DxO Optics Pro comes up with. DxO's always looks "flatter" and as a result more realistic to my eyes when viewing them. Thus, the differences are not (to me) insignificant at all.

My problem here DM is that images like those you show are of a kind where geometric distortion probably wouldn't be much of an issue to my eyes even if it remained uncorrected altogether. I usually find geometric distortion (typically barrel distortion) problematic only if there are objects toward the edges of the frame that are known to be straight but aren't.

Again, it comes down to personal perceptions here. Had I not compared DxO correction with other corrections, I would not be aware of the differences to my eyes. You may well see the same differences, and not consider them significant (or may yourself prefer the other one).

For similar reasons, I have difficulties quite understanding in this case even what you mean by expressions such as "flatter" (do you simply mean that lines that should be straight are straightened out to a greater extent?), ...

None of my examples are architectural photography. They are natural scenes where the eye has far less of any pre-conception of what something "should" look like. That can work both ways, then. It is indeed subjective. By "flatter" I mean less "raised towards the center" in areas nearer the center of the image-frame.

One cannot "take a measuring stick" to such natural scenes. DxO's corrections seem to please my eyes more - and (may) better comport with my visual memory. I cannot "prove" that - because my own visual memory (or anybody else's) does not read-out like a digital machine.

... "bulbous and bloated", or "extending corrections farther out" (isn't it necessary for distortion correction to have a pretty significant impact on the corners in order to be at all successful?).

The whole point of the Lightroom 3.41 compared with DxO Optics Pro 6.55 example was to show that LR 3.41's geometric correction was unable (despite my best efforts) have a "significant impact" upon the upper-left corner area of the (somewhat cropped) image-frame - whereas DxO Optics Pro 6.55's geometric correction was indeed able to do just that (without over-correcting the center area, as was the case when Lightroom 3.41 attempted the same).

Given that, I do not really understand quite why you pose your question (directly above). Of course, the answer would be "yes". The operational issue is, "when such an impact is made upon the edges/corners, what do the inner areas look like ?".  I think that DxO does use multiple reference-points, and in doing so is capable of implementing corrections that are more able to adjust outer areas without overdoing the corrections applied in the more central areas.

That is a factual matter. You argue that rectilinear distortion is simplistic in nature even in the case of multiple-element lens-systems. Another factual matter. I don't know for sure about that.

It would be interesting (though possible) that DxO Labs would create a geometric distortion correction system from characterizing lens-systems that is primarily related to human perception (and unrelated to their optical measurements). Anything is possible where it comes to "art" ?

What I do claim to "know" (and you in your own way claim to "know") is what seems to please our perceptual minds, and what we may also indeed perceive, but not concern our minds with. Those are not things that really can be litigated in "fact finding" endeavors. Thus, they are not things that can be declared as "important" or "unimportant" to the perceptual minds of others.

It seems that the best that we can do is to offer our own subjective perceptual preferences ?

Here is another LX3 wide-angle example for you. It involves crops that represent only part of the recorded image-frame, and subject matter that exists much closer to the camera lens. The two crops are not identical - so in viewing have a look at the general contour of the subject-matter.

LX3 in-camera distortion corrected JPG (cropped):

.

DxO Optics Pro corrected (cropped):

Do the same thing with these two Original images, and switch back and forth between them. A difference can be seen. Whether or not the aesthetics surrounding these visible differences constitute "meaningful" differences may be a matter of taste. The fact that one individual may prefer either one or the other correction method is valid - whether or not various metrics may or may not establish some kind of numerical "proof".

.

Then there exists an issue of corrections performed upon subject matter that exists far away from the center of the image-frame and nearer to the edges or corners. In the following case (a somehwat cropped DMC-FZ28 RW2 image), I was able to accomplish geometric corrections using DxO 6.55's manual controls (processing an 8-bit TIF) that Lightroom 3.41 was unable to accomplish in RAW processing mode (using the Panasonic image-file meta-data in conjunction with Lightroom 3.41's manual geometric distortion corrections).The differences seen are significant to my own eyes.

Lightroom 3.41:

Lightroom 3.41 RAW geometric distortion (auto + manual) corrections

Download the Original and look in the upper left corner of the image-frame in an image-viewer.

.

Now, download this DxO Optics Pro 6.55 manual geometric correction of an 8-bit TIF of the same image. It can be seen that DxO Optics Pro is capable of extending it's useful geometric corrections farther out into the image-frame than Lightroom (or Silkypix, or the OOC JPG, I find).

DxO Optics Pro 6.55 8-bit TIF manual geometric distortion corrections

I myself prefer the DxO Optics Pro geometric correction functional capability. Lightroom (your 4.x, anyway) does do some things (in my view) better than DxO Optics Pro - but not this. Of course, our perceptions are not immune from being shaped by what are our chosen preferred applications. And, subjective individual perceptions of any single performance element may also vary between viewers.

For more complicated forms of distortion, it is clear that more sophisticated methods are likely to yield noticeably better results. On the other hand, I wouldn't think it likely that the form of distortion generated by MFT lenses is as a rule very complex and haven't seen any sign of that with the MFT WAs I personally have/use. Complex forms of distortion are often the result of less than perfect attempts at optical correction and MFT lenses designed to be software-corrected for distortion are likely to avoid that particular problem.

Here is a geometrically uncorrected LX3 wide-angle RW2 processed using DxO Optics Pro 7.23:

Uncorrected

.

Here is the (in-camera corrected) Panasonic OOC JPG:

Panasonic LX3 corrected in-camera OCC JPG

.

Here is the best that I was able to do using RAW Therapee 4.0.9.184's manual geometric distortion corrections. It has a simliar contour as the Panasonic in-camera corrected OOC JPG - a contour that appears to have a single correction reference located at the center of the image-frame:

RAW Therapee 4.0.9.184 manually corrected

.

Download the Original versions of either of the above displayed corrected images and view them in an image-viewer, switching back and forth with the DxO corrected image displayed below. Note that because the LX3 in-camera OOC JPG is chopping-off a lot of the perimeter, the existing differences are easier to view using the (similar to the OOC JPG) RAW Therapee corrected image.

DxO Optics Pro 7.23 automatically corrected (normal 4:3 aspect-ratio)

The differences between the nature of the DxO correction and other methods can be readily seen, showing the "flatter" contour of the DxO (evidently multiple reference-point) correction.

.

Here is a somewhat wider-angle version using the same automatic DxO Optics Pro 7.23 geometric distortion corrections, taking advantage of it's ability to present a larger FOV (in cases where rectilinear barrel distortion exists in the recorded image). The resulting aspect-ratio is 1.392.

DxO Optics Pro 7.23 automatically corrected (wider 1.392 aspect-ratio)

DM ...

Reply   Reply with quote   Complain
Post (hide subjects)Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow