I swear I read something similar along the lines (pun intended) of this a couple years back but energy was not the Radon transformation, I forget what exactly it was. The hardest part of using this in production is that there is a lot of hand-tuned values, particularly during the edge detect portion which makes it difficult to scale. It's usually cheaper and easier to calibrate the camera at a mass scale in the factory using "old school methods.
I have been wanting to do this for my<p><a href="https://www.kandaovr.com/qoocam-ego" rel="nofollow">https://www.kandaovr.com/qoocam-ego</a><p>after reading Lenny Lipton's books about stereo cinematography I've been debugging my stereograms and one thing I know is that the lenses on that thing have a little bit of pincushion distortion which means stereo pairs that are supposed to be perfectly aligned vertically aren't quite.<p>I know DxO makes distortion correct filters for lens/camera pairs and I was sure I could make one by taking pictures of a grid but this gives a definite path to doing it.
I have difficulty understanding what the transformed image is equivalent to. This makes it feel like the picture was taken at a difference distance and focal length, but[1] it would look different if that were the case because the perspective would be different. Does this have any "physical" interpretation that would make it easier for me to understand? Like, cropping an image is equivalent to changing the focal length; what would this be equivalent to? A type of rectilinear lens?<p>[1] With the exception maybe for a single plane in focus?
His dissertation looks very interesting<p><a href="https://hh409.user.srcf.net/index.html#PhDThesis" rel="nofollow">https://hh409.user.srcf.net/index.html#PhDThesis</a>
This is cool, but couldn't you generate the correction transformation simply from knowing the lens geometry? I assume this is what my phone is doing when I take wide-angle pictures (which don't have any visible distortion)
A question for those who know optics: If the angle of incidence is past the critical angle of red do all of the visible spectrum get reflected without any chromatic effects ?<p>Are there cameras that have a sensors laid out on a curve matching the expected surface on which the image is in focus ?<p>I wonder why there are no cameras (apart from astronomical telescopes) that use reflection only for imaging. Such a camera would be too bulky to be practical ?
This site could really use a mobile version[0]<p>[0]: <a href="https://developer.mozilla.org/en-US/docs/Web/CSS/CSS_media_queries/Using_media_queries" rel="nofollow">https://developer.mozilla.org/en-US/docs/Web/CSS/CSS_media_q...</a>
It should be noted that this article talks about a pretty niche use case without really spelling it out.<p>Camera optics are generally designed <i>not</i> to exhibit this kind of distortion. As other commenters note, wide-angle lenses are ground to provide rectilinear projection where horizontal and vertical lines are straight. Further, if a particular lens does exhibit distortion, the usual solution is to measure the effect and construct a reverse mapping that can be applied in software.<p>There are relatively few situations where you have a distorted image taken with unknown lens, but where you have a regular grid of horizontal and vertical lines for the algorithm to rely on.
in-car racing cameras have very wide FOV. it's not uncommon to have such corrections applied to the video stream. i believe even the ubiquitous go-pro has such a filter.