Spheres in corner of frame, distortion, poor ptm

[mgts24]

A recent capture is producing a coarse, fuzzy PTM, even though the individual RAW shots were high quality.  I had trouble with the process of highlight capture because the spheres seemed to be out-of-round (distorted), and it was impossible center the guide circle on the image of the sphere.  I am assuming this led to a registration problem in the PTM, hence the poor quality.

These were shot on a 24–105 Canon L series zoom, on 60 mm, with the sphere right in the corner of the frame.  Am I right in thinking that: 1) zooms should be avoided and macros used where possible for their typically flat fields and lack of barrel distortion.  2)  the spheres should not be placed in the extreme corner of the frame but closer to the center on the edge.

 

Feedback welcome.

[marlin]

Hello mgts24!

 

thanks for the message. I will attempt to reply to your post with some useful information.

 

You have experienced Lens Distortion. The 24mm-105mm L is a really nice lens, but it may not be the best lens to collect RTI data sets.

 

As we know, Lens Distortion is in every lens -- thats just optics.  Some have more, some have less, a lot less. We steer professional towards the prime lenses.

 

Primes

The 50mm Macro Prime is one of our daily drivers in our camera bag:

http://www.bhphotovideo.com/c/product/12145-USA/Canon_2537A003_50mm_f_2_5_Compact_Macro.html

 

The other worker bee is the 100mm L (or its previous)

http://www.bhphotovideo.com/c/product/647011-USA/Canon_3554B002_EF_100mm_f_2_8L_Macro.html

http://www.bhphotovideo.com/c/product/194451-USA/Canon_4657A006_100mm_f_2_8_USM_Macro.html

 

Both of the above lenses have the least amount of lens distortion, or rather, the most acceptable amount of lens distortion.

 

Lens Resource

Websites like http://www.dpreview.com/ (go here and look up your lens), can offer valuable insight into the distortion(s) that are inherent of the lens in your camera bag.

 

For example check out this interactive chart on sharpness on the 100mm Macro L. There other good stuff in there too. (note the comments about the 100mm being exceptional in the distortion category)

http://www.dpreview.com/lensreviews/canon_100_2p8_is_usm_c16/4

 

Zooms.

A word about zoom lenses (lenses which are adjustable, e.g. : 100mm-400mm). Its not bad to collect RTI data with a zoom lens, but its a lot better to use a Prime lens --- we know this. But another consideration, is that zoomable lenses, lenses that have that turnable ring to adjust your FoV, will often times *shift - especially if the camera is pointed down towards the earth. Your very first image might be tack sharp and in focus, but then the last image in the RTI data set might be soft and out of focus. I've seen this numerous times. The lens simply shifted during the capture sequence. If you have to use a lens that translates to adjust the FoV, then use 'tape' (gaffers tape) to physically 'tape' the lens in place to itself ---- preventing movement and keeping all silent. (don't forget that you also Must be on Manual Focus "MF" mode).

 

Spheres in the corner.

Spheres in the corner of the frame will appear egg like with lenses that contain a lot of lens distortion. Even the best prime lenses might have a little bit of 'wobble' to them. Moving the spheres towards the middle of the frame even just a tiny bit can help the distortion. In some setups you can control the sphere placement, if you can do this, you might as well. Other times, bc of the subject the sphere may end up in the corner. Remember that RTI Builder is pretty flexible and will most times, more than not, still offer a usable and acceptable result ---- even if the sphere is egg like. (hopefully enough to reveal that hidden text you are looking for).

 

Lens Distortion Correction.

quick word on Lens Distortion correction. If you want to correct for lens distortion, there are a few ways to do this. Its too much to disscuss in this post but, if you do correct your images, you would want to correct them at the DNG level or at the 'jpeg-exports' phase, and then continue to process your data from there.

 

Hope that this Helps.

Happy F-stop.

 

Marlin.

[mgts24]

Thanks Marlin,

 

I ran another series of shots of the same subject but with my 50 mm f2.5 macro, left more space around the subject, and with the sphere half way up the side of the frame and closer to the center.  What a difference!  The sphere was nicely round, and all the fuzziness of the previous capture was gone.  I will have to see if there's room in the budget for the Canon L series 100 mm macro, or the Tamron 90mm, which is also very good.

[Taylor]

Thanks MarIin and mgts24 for an interesting post!  I vote for another post on lens distortion correction.  For my particular camera (modified Panasonic Lumix GH2), Lightroom 4 doesn't provide any generic lens correction profiles.  From reading dpreview, it appears that many (if not most?) digital cameras apply lens correction parameters in the camera's firmware.  Some lenses are quite good with or without the firmware corrections; others, not so much. 

 

It would be interesting to know how these lens corrections, under standard operating conditions, affect the accuracy of normals.  Can the corrections in the camera's firmware be improved by generic profiles in Adobe products (if they're available)?  My guess is that the way RTIs are captured and processed, there's an intrinsic correction as long as the RTIBuilder or PTMBuilder software can find the center of the highlight on the spheres, and as long as the spheres aren't too distorted, the errors should be small.  Can generic profiles be improved upon for a particular (non-generic) lens, and what's the best way to correct distortions that are unique to each lens in the RTI process pipeline?   Then you've got your orthorectification, flat-fielding, vignetting, anti-aliasing, sharpening, noise reduction algorithms, and occasional dead pixels, etc. to correct for.

[cdschroer]

I have a couple of thoughts about this.  First, I will say that I think you are seing 2 different things going on.

 

1.  Yes, we do indeed recommend prime lenses (non-zoom single focal length) for RTI whenever possible

 

2.  The fact that the spheres are ovoid is to be expected due to barrel distortion in your scenario.  This will not affect the outcome of your RTI, in terms of it's visual quality in the viewer.  The highlights on the spheres are used to determine the light position for each image.  The light position is critical to accurately calculating the surface normal. (see http://culturalheritageimaging.org/Technologies/RTI/#how_RTI for a description of surface normals)  However, you will not visually see a difference in the surface normal due to inaccurate light positions.  This can affect your data for other kinds of uses.

 

2(a)  You have a couple of options in this scenario to get a better sphere.  You could distortion correct your images before bringing them into RTiBuilder.  OR you could just make sure that the center of the sphere in RTI builder is in the center of the ovoid, even though it will be cutting off some of the sphere, since the sphere isn't round.  This isn't perfect, but it will give you the best possible result with this issue.

 

3.  The issue you describe of having a fuzzy result when the input images are good is likely caused by movement within the image set.  This could be from the camera or the subject moving slightly or since you were using a zoom lens, through the focal length shifting slightly across the whole image set.

 

There isn't a great way to align an image set that isn't labor intensive.  You may want to reshoot the subject, if that is easy.  If not, there are ways to align large image sets, but they aren't fun.  There is a tool specifically for doing this for RTI data sets, developed by a team we are working with at Princeton, however it is still in a Beta state and not yet released.

 

Carla

[cdschroer]

I'll note that I hadn't seen the additional discussion here when I posted the above.  (had a tab in my browser from this morning and didn't refresh)  I hope I still added something useful to the discussion, though clearly some of the issues had already been discussed.

 

As for distortion correction - there are different ways to do it.  Some are better and more accurate than others - specifically doing a distortion correction based on a photgrammetric calibration sequence for your specifc camera lens, and focus distance is going to be the most accurate.  Doing this on an RTI image set should give a more accurate resulting RTI.

 

Even using a database type distortion correction, such as comes in Photoshop where you pick your camera and lens can give you a better image than one with a lot of barrel distortion.  If you are using the RTI in the viewer for visual inspection, the difference you wil notice is that the subject isn't being shown with this distortion.  The fact of the spheres being distorted won't affect your viewing experience, unless you don't crop them out.

 

Where more accurate light positions matter is with other kinds of uses of the RGB + Normals (RGBN) data such as to track change, or to use within algorithmic rendering. 

 

Carla