Resizing Algorithms compared
This is a comparison of three resizing algorithms.
Link to full size: www.flickr.com/photos/kinematic/13783164545/sizes/o/
The first is the common method of Bicubic resizing (200%) with unsharp mask of 3 pixels radius applied.
The second option is my own custom workflow of 4 algorithms combined with super resolution (which I call resolution +) and no sharpening.
The third is one used by programs like Perfect Resize and many other popular resizing tools, Lanczos resizing algorithm with RL Deconvolution sharpening.
The Lanczos is definitely an improvement over the Bicubic method, giving a much crisper cleaner resize. However RL Deconvolution does tend to over-sharpen things a touch. I might even say that it artifically sharpens things that shouldn't be sharpened (note the little white sharpened specks on the rock).
My hybrid method is still better suited reducing essentially diffraction issues and recovering details. Especially viewed at 100%, it appears not just as a resizing of the image by 200% but an actual increase in resolution without the artifacts that seen in the other two methods. Although it is a little labour intensive, with RAWTherapee workflow the hybrid method definitely yields favourable results.
It is not something that one might want to do on every image, however the effort is worth it for those times that you either are resizing images for large reproductions or you're cropping heavily and need to recover some details.
Experiments like this certainly disuade me from ever needing to pick up an A7r for more resolution in studio situations (which most of my clients these days rarely ever call for or need).
I'm very close to publishing my workflow and description on how this is done. The cost of software is reasonable and the technical aspects behind it have been far simplified with the introduction of RAWTherapee into the workflow.