Basically it is a single pass filter using a circular (or elliptical) weighted sampling of the pixel neighbourhood, instead of a 2 pass linear sampling. It is used for image distortions where rotations can be involved (as well as scale).
From IM Examples...
The Lanczos equivalent would thus be a Bessel windowed Bessel filter.The Jinc() function (more commonly known as a 'Bessel' filter) is the equivalent filter for use in a two dimensional cylindrical, or radial filtering operation, though it is not a perfect fit in this regard. Though very similar and closely related to Sinc() (see the graph provided) it is designed to filter a rectangular array of values using a radial or cylindrical distance, rather than orthogonal (axis aligned) distance. As such the first 'zero-crossing' falls between the values of 1.0 (for orthogonal neighbours) and the square root of 2. That is it crosses zero at approximatally the value of 1.2196, so as to match as close as posible all its 'close' neighbours.
I have seen on paper that did use this specific variant, and the improvement in a rotation distorted image, exampled in the paper, was quite simply amazing. However I have not reached the point of actually duplicating those results.
Hmmm... Can't seem to find the reference to the paper in my bookmarks.
Perhaps I have a physical copy. I'll have a look.
Actually I am more interested in a Mitchell like cylindrical varient.