Do Leitz use spherical aberration before after stop to deblur and increase resolution

[Mustafa Umut Sarac]

I found an Ernst Leitz Canada ELCAN paper on Leitz Summicron 35mm and interestingly there are highest amount of aberrations before and after the stop .
I am talking about two lens elements.

Highest aberration was Spherical Aberration 3.

Let me introduce the strangers to computer image correction world. For last 10 years , there is a increasing amount of papers discusses the image correction.
And after 2010 , there is a trend to put a single element or very poor quality lens which has so much aberrations but MOSTLY SA3.

They say if there is spherical aberration very strongly presented in lens , its very good thing to correct everything to unlimited amounts compared to lenses.

They say , if there is SA at a lens , correction error becomes too low to afraid and you can deblur limitless.

I found that Leitz lenses are small diameter , their glasses %20 higher refractive index but their resolution is 4 times of average lens.

I thought , they use SA3 before the stop and after the stop , they deblur limitless to make resolution fly.

May be I am wrong.

What do you think ?

Mustafa Umut Sarac
Istanbul

 

[Nodda Duma]

Do Leitz use spherical aberration before after stop to deblur and increase re...

Hi Mustafa,

If I'm following you correctly, the theory is to design an imaging lens as part of a system, and move some of the aberration correction into post-processing (performed by camera electronics). The lens corrects everything but spherical aberration, which is a simpler design problem. The amount of spherical aberration is known, and used to generate the point spread function of the lens. Then, when the digital image is captured, spherical aberration is corrected in post-processing by deconvolution of the image with the point spread function. The only issue with leaving SA3 in corrected is that it looks a lot like defocus from a design standpoint. The result would be significant image artifacts for out of focus parts of the scene.


This is similar in theory to the technique used by smart phone cameras to generate such nice images out of such a small lens....although the optic is design with a more unique point spread function for reasons described above. It's a hot spot of research in optical design, because there are significant gains to be made in simplifying lens designs. Imagine an optic with Summicron-like performance but only using three pieces of glass. Look up "computational imaging".



For the classic Summicron design (including the design you describe above), this isn't what's going on. Obviously you can't post-process for a fine-looking negative. The Summicron is a double gauss, with SA3 traditionally found as you describe. Correcting it isn't a very difficult problem to solve. For a fast double gauss, there is a lot of residual SA3 present which you have seen. You would think the imagery doesn't look good but the solution is to use higher order spherical aberration (which doesn't show up in your plots) to balance out SA3.

Odd aberrations like coma and distortion are corrected perfectly in a double gauss by symmetry of the design about the stop. Higher-index glass (and lower dispersion) helps correct color and reduces steepness of surface curvature on the lenses which helps tolerances in assembly too.

Btw can you link to the paper? It'd be interesting to read.

 

[Mustafa Umut Sarac]

Double Gauss lens design: a review of some classics Reginald P. Jonas M. D. Thorpe

Double Gauss lens design: a review of some classics
Reginald P. Jonas and Michael D. Thorpe
ELCAN Optical Technologies, 450 Leitz Rd, Midland, Ontario, Canada, L4R 5B8

 

[Mustafa Umut Sarac]

I am asking does post processing done in the lens when thinking lens as a optical computer ?

Umut

 

[Nodda Duma]

Ah gotcha. I wouldn't think of it that way...aberration correction and optical computing are two different things entirely: their only relationship is that they use lenses.

Optical computing takes advantage of the fact that an image plane is the (2-dimensional) Fourier transform of the pupil of an imaging system. So, if you want to perform pattern recognition on images (like finding the letter 'F' on a sheet of paper), you can take advantage of this transformational property to perform massively parallel processing.

Of course, this was more interesting 20-30 years ago when computers were not as powerful. Now it's very simple to run a matching algorithm in Matlab (for example)....significantly faster and more portable than setting up an optical computer in the lab.

Spherical aberration is simply an error in this transformation process, much like static noise on an electrical signal. For best optical computing results (similar to electronics), the optics you use should minimize that and the other aberrations. Aberrations don't even "factor in the equation" so-to-speak. In light of this, and the fact it's a linear system, transformations inherent to an imaging objective cannot be broken down further into the individual element contribution. Much like trying to describe the performance of a desktop computer in terms of the copper wire thickness. Kind of doesn't make sense.

 

[Mustafa Umut Sarac]

When you know about the technology , please answer this.

How old Leitz lenses resolve 250 lines but the new zeiss lenses 450 lines per milimeter ? Either lenses have nearly zero aberrations , which factor contribute to resolution ?

 

[Nodda Duma]

Do Leitz use spherical aberration before after stop to deblur and increase re...

Computer aided optimization, tighter fabrication and assembly tolerances, and better selection of optical glass all contribute to performance improvements of modern lenses.

In short, lens design is 30% theoretical and 70% practical considerations.

 

[andrew.roos]

Thanks, interesting paper. Since I work a bit with design optimization, I was interested in the finding that computer optimization did not significantly improve the 35mm Summilux, which is testimony to Mandler's exceptional ability. Of course for companies without a Mandler, computer optimization may have had a greater effect - here I'm thinking for example of Bronica, where the "P" series lenses, which I believe were computer optimized, are in several cases significantly better than the previous designs.

 

[alanrockwood]

Hi Mustafa,

If I'm following you correctly, the theory is to design an imaging lens as part of a system, and move some of the aberration correction into post-processing (performed by camera electronics). The lens corrects everything but spherical aberration, which is a simpler design problem. The amount of spherical aberration is known, and used to generate the point spread function of the lens. Then, when the digital image is captured, spherical aberration is corrected in post-processing by deconvolution of the image with the point spread function. The only issue with leaving SA3 in corrected is that it looks a lot like defocus from a design standpoint. The result would be significant image artifacts for out of focus parts of the scene...

Another problem is that deconvolution schemes almost always amplify noise. The reason is that much of the noise in a signal resides in the high frequency components. In the case of an image this would be spatial frequency. Deconvolution bumps up the high frequency components in order to sharpen the features, but noise gets amplified along the way.

Also, in practice the improvement in the resolution of the signal obtained by deconvolution tends to be modest. Trying to go too far ends up introducing distortions due to amplification of small poorly characterized errors in the system.

 

[Nodda Duma]

Do Leitz use spherical aberration before after stop to deblur and increase re...

Yep. Sometimes you can accommodate the increased noise depending on the application. If not then you either constrain its use to relatively noiseless conditions (ie have enough bright light), or you get into blind convolution techniques and /or tailoring the PSF for each individual system. A real problem (again depending on the application) is trying to do all this real-time at a useful frame rate with very constrained SWaP requirements. The additional "tuning" of the fundamental concept can be processor intensive.