Piccure+; Yet another sharpening solution

[revi]

(...)

One thing that has become clear to me over the course of evaluating the various tools. When properly applied the deconvolution method is superior for correcting optical aberrations than traditional USM or high pass sharpening. (...)

That is because deconvolution methods actually try to correct the abberations (i.e. increasing resolution). Mathematically speaking, optical abberations correspond to a convolution of the image with a specific function. If you know that function, you can use it to remove the abberations from the image, and that's what deconvolution is trying to do. "Trying to do", because the exact functions that describe the abberation are rarely known. Approximations (or ideal versions) are known, so those get used.
Results can be very spectacular when looking at 100% size, but overdoing this kind of sharpening tends to give very ugly artifacts (.

And it's a very time-consuming method compared to USM: typically, deconvolution consists of an inverse Fourier transform, multiplication with the inverse functions and a forward Fourier transform (Fourier uses sine and cosine base functions, other methods, like wavelets, use different base functions, but the principle stays the same), and that's a lot of calculations...

Also, the effects are often not all that spectacular on a reduced image: the fine details you want to get out through deconvolution become invisible when you reduce an image to 1/4-1/10 of its original size. (not at all impossible: 20MP images at 2:3 format are about 5500 pixels on the long side, 1:7 reduction would give you an image of 780 pixels long)

Unsharp masking and related methods do not correct any abberations, but increase the contrast at the edges in the image (increasing acutance).

Remco

[NorthernFocus]

That's a nice summary explanation, Remco. Most of the tools which claim to be apply deconvolution methods are obviously gross simplifications simply due to the fact that they arrive at corrections almost instantly. Piccure is either doing a lot more calculations and/or is poorly (i.e.inefficient) written code based on the time required to arrive at a solution. And for the reasons you state regarding downsampling, it is not for general purpose sharpening.

My only interest is due to the fact that I produce a lot of large prints. With some of my older images captured with older, lower resolution cameras I often find myself enlarging rather than downsampling images. Also I have some older images with excellent content that were captured with not so good glass and with cameras that had AA filters on the sensor. Where the Piccure looks to be really useful is in those cases. Applying the deconvolution as "capture" sharpening prior to enlargement and then applying traditional sharpening methods for output sharpening has worked well on a couple of them.

When I get around to it I'll post a couple of examples.