What is mFT FoV?

Started Jan 27, 2013 | Questions thread
Shop cameras & lenses ▾
Detail Man
Detail Man Forum Pro • Posts: 16,400
Re: What is mFT FoV ?

gollywop wrote:

Detail Man wrote:

Detail Man wrote:

gollywop wrote:

If one has, say, a 12mm lens on an mFT camera, it presumably implies some FoV. However, these days a large number of mFT lenses do software correction for lens distortions, and the image as rendered in, say, ACR is significantly different from the uncorrected version of the same image.

My question: is the 12mm FoV relevant to the FoV of the original image or the FoV of the corrected image, which is, typically, significantly smaller than that of the original?

I agree that the diagonal FOV (thus the 35mm equivalent focal length when the crop factor is considered) of the default (rectilinear distortion corrrected) 4608x3456 pixel size is designed to be relatively close to the intended specification.

There do exist additional photo-sites on M43 image-sensors. They are in the horizontal sections. Some of those photo-sites are unusable - because they are masked-off, and used in order to report black-level offset information only. There are only a few extra photo-sites in the vertical dimension (but they are apparently set aside for use in implementing rectilinear distortion corrections).

The image-file meta-data for your E-M5 (4:3 AR) only reports "Image Dimensions" the calculated diagonal FOV of which is only very marginally different from the default value of 4608x3456. Those pixel-dimensions (for the 3.73 Micron pixel-pitch of the E-M5's image-sensor) calculate out to be only very marginally "longer" (in 35mm equivalent focal length), meaning "narrower" (in FOV) than the specification value.

See: http://forums.dpreview.com/forums/post/50738679

However, it is clear (from your posted example as well as another poster's example) that the E-M5 image-sensor does have some extra (horizontal) photo-sites that are not reported in the E-M5's image-file meta-data.

GH2 RW2 (4:3 AR) image-file meta-data reports "Sensor Dimensions" of 4760x3472.

See: http://forums.dpreview.com/forums/post/50738731

DxO Optics Pro 7.23 (only when automatic distortion corrections are enabled, and "Keep Aspect Ratio" is unchecked) widens the horizontal field (only), and produces output images that are 4668x3457 (as compared to it's default output image pixel-size of 4609x3457). The wider horizontal pixel-dimension yields a diagonal dimension that is 0.821% larger than default diagonal pixel-dimension.

Have been playing around (using RAW Therapee with the same GH2 RW2 image-file used with the DxO Optics Pro experiment (above). It allows a maximum of 4616x3464 pixel-dimensions (with or without rectilinear distortion corrections being changed in value), and with or without "Auto fill" of the frame (so that no sign of blackened areas outside of image-data being used shows).

It appears (I think) that both DxO and RT are (in the course previewing in-process images) are re-sampling the original de-mosaiced image-data in order to cause them to comport with maximum processed image pixel-dimensions that the application developers have chosen for the particular camera model (with or without rectilinear distortion corrections being applied).

Yes, this one has had me baffled for some time. I noticed with my Canon s90, which was the first camera I had that did software lens correction, that although the rectilinear distortion of the lens varies by focal length, and therefore the correction must also vary with focal length, the demosaiced uncorrected raw files are all the same size (3744x2784) and the demosaiced corrected files are all the same size (3648x2736) regardless of focal length.

I can understand how the uncorrected images can all be of the same size, but it seems to me that the corrected images can only be of a same size if some resampling is going on, different amounts for the different corrections at different focal lengths.

Indeed, rectilinear-distortion correction processes are (necessarily) resampling pixel-data (by varying amounts, depending on their location in the image-frame). Nothing pristine about it, sad to say.

While it may well be the case that camera manufacturers are attempting to ensure that their distortion-corrected in-camera JPGs are (typically) providing a diagonally referenced 35mm equivalent focal length (corresponding to a diagonally referenced Angle of View) - at least where it comes to their own manufactured lenses, anyway - that is something close to the rated AOVs of (their) lenses ...

... it appears that theses RAW processors (at least those not provided by the manufacturer for use with their own camera models, anyway) are (by re-sampling the original de-mosaiced image-data) taking liberties based on their own independent judgments of how many extra image-sensor photo-sites can (safely) be utilized in the case of image-sensors on any particular camera model without running into "masked" photo-site locations on the horizontal egdes, and without running into photo-site locations needed for rectilinear-distortion corrections on the vertical edges.

DxO Optics Pro does not use the rectilinear-distortion correction data that exists in RAW image-file meta-data. They use their own data derived from characterizing camera/lens combinations.

RAW Therapee appears to utilize the rectilinear-distortion correction data that exists in RAW image-file meta-data (in the case of their experimental "Auto Distortion Correction" button that works in the case of some camera models only, the DMC-GH2 included), but that is an optional control existing outside of their own manual rectilinear-distortion correction tool.

Among my converters, ACR uses the meta-data, but both PN and RPP do not, and they both produce identically sized and shaped uncorrected images. None of them allows for any options: they either correct or they don't.

How much larger (in pixel-dimensions, relative to 4608x3456) are the ouptut images from them ?

Due to optical differences that exist between various lenses, as well as the re-sampling of the original de-mosaiced image-sensor (even in cases where rectilinear distortion corrrection is not being applied at all), it seems that one is not able to accurately predict the Angle of View (hence, the 35mm equivalent Focal Length) by performing calculations based upon known pixel-pitch specifications of the camera's image-sensor and the expected to exist active photo-site dimensions.

Further, the lens-systems are multi-element (and in the case of telephoto lenses, asymmetrical with a Pupillary Magnification value less than unity), and exhibit "focus breathing" and effects which make the actual AOV hard to predict for any given Focal Length and Focus Distance (as I vaguely understand such things).

The commonly used formula:

AOV = ( 2 ) x Arctangent ( ( Frame Diagonal ) / ( ( 2 ) x ( Effective Focal Length ) ) )

... descibes a single, symmetrical lens only.

An excellent reference: http://www.panohelp.com/thinlensformula.html

That does indeed complicate matters.

Thus, as Anders has referenced, using the following emprical method seem the only approach:


while also using a specific RAW processor configured to yield the widest (usable) observable AOV.

Well, I'm really interested in what the AoV for a given focal length is after correction, and whether that corresponds to the AoV that should be associated with that focal length.

Perhaps you should instead adopt the viewpoint of some prominent physical scientists, and satisfy yourself with a more practical, and thus more earthy, robust, and virtuous 10% error tolerance ?

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