10.5mm CA correction

[ultimind]

I've noticed all of the threads recently about the 10.5mm and how much everyone seems to like it. I know I love mine to death. But the chromatic aberration that this lens seems to have is absolutely terrible in my (un-professional) opinion.

Are there any options short of buying DxO that'll fix this? I haven't had much success in Photoshop CS3 (Camera Raw 4, or Lightroom for that matter) fixing it.

Here's a prime example... Look at the trees to the right. Photoshop can only correct this to a minor degree:

Subscribe to see EXIF info for this image (if available)

 

[ultimind]

I love this lens for landscapes and I often find fine detail at the edges of the frame. I never unwrapped the shot above, but it gets many times worse when you run the image through a rectilinear converison. Hemi conversion keeps it pretty much in tact as it is.

 

[weiran]

DxO might be a worthy investment for this lens, it can even defish the shots for you.

Nice photo by the way.

 

[Muonic]

I've had pretty good luck using Nikon Capture 4 to correct CA from the fish. Just click one button, and you're done.:smile:

 

[pivlrs]

resolution is preserved

Personally I would not mess with your image with any kind of conversion - it does not need it :smile: as to the two types of conversion I find rectilinear conversion to be totally useless due to the stretching of the image and loss of resolution JMO

Although I agree that I (usually) don't like the extreme expansion of areas at the edges/corners resulting from rectilinear mapping, I have one nit to pick on your complaint above...

the "resolution" of the image is the same in either case - i.e., the amount of detail resolved in the scene is preserved regardless of mapping to other representations.

perhaps a better way to describe what you are saying is "exaggeration of image defects in the corners" or something...

-matt

 

[Tri-Elmar]

I was always trying to love the results from this lens so I could justify buying it... I never did.
To be sincere, this thread didn't help either. :667:

 

[ultimind]

The 10.5mm is definitely a special purpose lens, and we have to live with it's shortcomings if we want it's focal length and distortion. I bought it for extreme sport photography because it's literally the ONLY lens available to do this with.

Some people cannot have a shortcoming in a $600 optic and that's understandable, but it's not an everyday lens like a 28-70 or 80-200 would be.

 

[pivlrs]

First here is an un-cropped rectilinear conversion of a fisheye image;

Subscribe to see EXIF info for this image (if available)

Now here is the cropped image that you would end up with using Nikon Capture:

Subscribe to see EXIF info for this image (if available)

and finally this is all of the information that was in the original fisheye image that has been discarded:

Subscribe to see EXIF info for this image (if available)

This is why I say that rectilinear conversion causes a huge loss in resolution. As you can see the converted image has lost a huge number of pixels that were in the original image.

I agree with your general description of what happens in a rectilinear conversion, and agree with you that the cylindrical projection of fish-eye images is usually the more pleasant and useful thing to do.

I simply disagree with what is meant by "resolution". "resolution" is generally a way to describe how much detail in the SCENE is discernible. That's the whole point of "resolution targets" after all. This property is preserved regardless of geometrical transformations of the image. In the example you use, the dominant effect of the rectilinear transformation is that some regions of the image are magnified in unflattering ways, allowing you to see the blur/CA more clearly (generally not a good thing!).

Put another way, imagine that there was a ruler at the edge/corner of the scene. if you were to pixel-peep at the original fish-eye image, you may say something like "hmmm... according to the ruler in the scene, i've got about 2.3 mm of blur". Now transform the image using any geometrical mapping you choose, and repeat the exercise. You should always see "2.3 mm of blur". I.e., the resolution is the same.

In the extreme, it would be possible to devise a mapping that spread that blur across the entire final image. The result would look crappy, but the "resolution" of the image would be the same as the original.

-matt

 

[ultimind]

I would define resolution as the camera and lens' ability to resolve detail. In rectilinear conversion, you are cropping the image but if you don't enlarge that cropped image, the resolution is the same. If you were to enlarge that image back to say.. 3008x2000 if you're shooting on a D70, 50 or 40... then your resolution for the target image size of 3008x2000 will be low than it would be if you left the image uncropped at whatever size it becomes after the geometric conversion and crop.

It kinda gets into the print world with this idea... I can have a 320x240 or a 3008x2000 image that both show (or resolve) 1000 lines of resolution, but since the 320x240 image would have been cropped... in order to retain that resolution on paper, I would have to print it much smaller than the 3k x 2k image that has the same amount of resolution (detail).

 

[pivlrs]

when I think of resolution I think of the total number of pixels in the image. if in rectilinear conversion some of the pixels are discarded then wouldn't the image now have less resolution?

If we forget about the stretching of the image we still have to deal with the fact that approximately 1/3 of the original pixels have been discarded. if you now print to the same size how can you not have less resolution?

As I said before, it all comes down to definitions. I have always thought of "resolution" as a property of the optical system, rather than a property of the presentation of the results. With digital, it's a tad more complicated, since imaging system is more than just the lens, but the concept is the same. This definition is shared by, e.g., the astronomy community, which typically defines resolving power in terms of how far apart do stars have to be (in arc-seconds) in order to determine there are two stars there versus one.

Of course, this definition requires that if you transform your image, your yard-stick used for measuring must also be transformed with the same mapping.

You seem to be working with a definition of resolution based on something like "sensor pixels per inch on final image". That's fine, in that it describes the problems with rectilinear mappings of fish-eyes. However, I think it's probably a somewhat unusual definition of "resolution".


Your meaning is clear however. And your recommended PP is, imho, correct.

-matt

 

[ultimind]

I'm not even sure that there is a proper mapping solution for fisheye images that would allow you to keep all of the original image in tact.

Since our displays (and prints for that matter) are a flat plane, we're limited in our ability to properly display such a wide field of view without distorting it. Geometrically, the rectilinear conversion is the most correct but not the most pleasant to the eye. The Hemispherical mapping of the Hemi plugin allows for a much more pleasing image provided you've photographed people. I think it would require a curved viewing surface to properly map a fisheye without distortion.

This could be an interesting experiment...