1 person found this helpful
The effect synthesizes a vector field. Multiple regions provide a means to modulate that field, i.e. the vectors' magnitudes and directions. What the underlying formula actualy is, nobody will tell you, but you could just assume that it's complicated math and probably involves things like Fourier transforms or other such stuff and that the "blending" of the vectors is done on the transformed function results, then converted back to the pixel data. It's definitely more complicated than just overlaying different regions with falloffs/ feathered edges, thouzgh of course there will have to be something similar to avoid visual artifacts when applying the result...
Thanks for the response. I think I'm reasonably comfortable with what's going on under the hood, but I'm still a little confused as to how the tool will work with multiple blur traces. I guess as a thought experiment I imagine the situation where the image has experienced rotational or arcing motion. If the calculations exist in the spatial frequency domain (Fourier space), then a circular deconvolution won't really be possible. In fact any blur with rotational symetry should be problemmatic to reverse (another example would be a radial blur) unless the image gets resampled in a non-rectangular fashion. So let's say there is an image that was slightly rotated so that the left hand side of the image experiences an arc shaped blur that looks like the left half of a circle, and the right side an arc that looks like the right side of a circle. The size of these arcs will increase with distance from the center of rotation. Is this simply a hopeless case where the blur trace that fixes the right side of the image negatively impacts the left side an visa versa, or is there some additional set of smarts at play when multiple blur traces are utilized? Is there a correlation function employed to optimize where a particular blur trace is used to deconvolve the blur function? Academically I'm interested in the algorithmic answer to that question, but from a more practical standpoint, I'm simply looking at how I use these multiple blur traces. Do I duplicate the layer apply one blur trace to one layer and another to another layer and then layer mask them together? If that's really the way to go, then what is the purpose of multiple blur traces? I guess I'm just really curious in the end. Thanks again for the response.
But you forget that motion blur includes the colors of any object behind the blur trail, allowing for reconstruction of contrast and color information, and that under the assumption that a physical camera was involved, the lengths of the trails are determined by the shutter duration, sensor delay and otehr parameters, which could be calculated based on arbitrary sampling data from the image - blurred or unblurred alike. Likewise, you would figure in teh lens curvature to determine the actual blue amounts. Of course you are right that certain images just won't be possible to reconstruct, but that's not only true for radial blur artifacts. As to you specific usage question, you wouldn't duplicate anything - unless you want to manualyl control everything. Just like content aware fill, the effect generates a seam blending map and does all that itself. Also using multiple layers may have the adverse effect of impairing consistency of the synthesized variants...