Can computational correction with simple lenses really compete with conventional lens design?
Asked 7/5/2014
4 views
2 answers
0
A paper on "High-Quality Computational Imaging Through Simple Lenses" suggests using simpler optics and then correcting lens artifacts computationally via the lens point spread function (PSF). In theory, this could replace some of the complexity of traditional lens design.
Can this approach realistically match conventional lenses in real-world photography, especially for slower/cheaper lenses? Or is the main limitation that deconvolution works poorly once sensor noise is included, making it more of an academic idea than a practical replacement?
Originally by Photography Stack Exchange contributor. Source · Licensed CC BY-SA 4.0
Photography Stack Exchange contributor
12y ago
2 Answers
2
The problem with any attempt to reverse optical blurring by estimating/modelling the point spread function is noise. In principal if you know how the lens blurs an image and have an accurate version of the blurred image you can reconstruct the original "unblurred" image.
But in the presence of noise you don't really have the blurred image, you have the blurred image + a load of uncertainty that you can't remove, and deconvolution methods fall apart very quickly.
I can't imagine there's much of a market for lenses that can't be used above ISO100...
Originally by user1375. Source · Licensed CC BY-SA 4.0
user1375
11y ago
0
Generated from our catalog & community — verify before relying on it.
In principle, yes: if you know exactly how a lens blurs the image, you can try to reverse that blur computationally.
The practical problem is noise. Real images are not just “blurred”; they are blurred plus sensor/read/photon noise. Deconvolution tends to amplify that noise, so image quality can break down quickly unless the capture is very clean. That makes this kind of approach much less robust in normal shooting than simply using better-corrected optics in the first place.
So the idea has real substance mathematically, but its usefulness is limited by signal-to-noise ratio. It may work better in controlled situations or at very low ISO, but it is not generally a drop-in replacement for conventional lens design. That helps explain why traditional optical correction is still dominant, even for cheaper lenses: optical correction avoids having to recover detail from noisy data after the fact.
Recommended products
UniqueBot
AI12y ago