This answer to another question of mine linked to an interesting article: High-Quality Computational Imaging Through Simple Lenses. It proposes the use of simple optics and computational photography techniques to compensate for the different artefacts that arise instead of the complex lens systems we are used to today.

I'm an engineer and do understand the mathematics behind the paper, but I seriously doubt that the designers behind the far more complex commercial lenses haven't given it a thought (and have a strong reason not to implement it). I understand that the nature of the PSF (point spread function) introduces problems at wider apertures, but there are cheaper slower lenses for DSLRs today that could use this technology. If it was a viable alternative it should already exist.

Of course the introduction of these lenses if they could compete with conventional lens systems would kill the manufacturers own market of cheaper conventional lenses, but it would also give them an edge towards the competition. There's also the (what I think is very slim) chance that the designers haven't thought about it. It can also be as simple as that the method doesn't deliver the quality that complex lens systems do.

Has this method any real substance to it and a real world application or is it just wishful thinking from a very academic point of view?

Note that I'm not picking on the scientists behind the paper in any way. New ideas are great and great discoveries are made all the time, but a lot of research never makes it to the industry.

  • 3
    \$\begingroup\$ I understand the idea of what is going on using this technique. (I wrote a prototype of such a thing myself). I'm not 100% sure on how to implement deconvolutions fast, and how fast they are. But I think that the computational power, needed is huge. Also, I guess that the detail in the photo has to be very high. My prototype did not work using 8-bit PNG images. I needed to use high detail colors (like 16 bit) to get the desired results. Just as a demo, the way I implemented the deconvolution (which was with a huge system of equations), my computer computed for over 5 hours for a 0.3 MP image. \$\endgroup\$ Jul 5, 2014 at 13:51
  • \$\begingroup\$ Oh, fortune-telling! Can I play too? The technique relies on knowing everything about the lens at all apertures and focal distances, so there's no point looking for it on an interchangeable-lens system unless it's a walled garden. And it transfers the cost rather than eliminating it. So howzabout using it to make fixed-lens compact cameras more compact? Sony's curved sensor is aimed squarely at the same target. \$\endgroup\$
    – user28116
    Jul 5, 2014 at 15:50
  • 1
    \$\begingroup\$ I'm not sure the community would be able to provide any kind of canonical answer to this beyond "time will tell." My take is that such algorithms will improve in efficiency and increases in processing power (including dedicated hardware/asics) to process in a reasonable enough time and are more likely to become part of a post-process workflow on the basis that if they can provide significant improvements to 'poor' or simple lens designs then more complex/heavily engineered glass ought to benefit also. \$\endgroup\$ Jul 5, 2014 at 15:51

1 Answer 1


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...


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.