Conceptual Software Produces Amazing Detail From Bad Lenses

The standard motto regarding photography equipment is "invest in glass" with good reason. While sensor technology certainly has a tremendous impact on the quality of the photograph, the real limitation often lies in the optics. One need only look at the difference between a kit lens and a top-level prime to prove this. However, new methods of image processing may soon alter the effectiveness of sub par glass. Researchers at the University of British Columbia have made some serious headway in a project they call "High-Quality Computational Imaging Through Simple Lenses." The project in it's current form uses simple lenses, meaning a lens with a single glass lens element rather than the complex multitude of optical elements in a standard camera lens. This means that these test lenses suffer from problems that complex lens systems have either lessened or completely eliminated such as chromatic aberrations, geometric distortion, and field-curvature to name a few.

Fixing these problems post-image capture is something that even highly advanced raw processing algorithms still struggle to perfect from amazing lenses. This team is showing that it can be done effectively using what equates to cave-man photo gear.

The result speak for themselves and are obviously pretty exciting when looking into the future for what this could be developed into. If this can be done with horrible lenses, imagine what could be done with just average optics.

Turns out they've started to answer that as well. The images below were shot with a Canon 28-105:

Original
cap
Processed
cap_deblurred

If you ask me that's pretty incredible given that it's still very much a work in progress. As you probably heard in the video there are still quite a few limitations. Even still, I would imagine that a certain software company will be snatching these developers up very soon.

Via Reddit

Log in or register to post comments

26 Comments

Chris D Johnson's picture

So what I got from that was:
1) These guys are way smarter than me.
2) Where do I give them my money?

Joel Quinn Photography's picture

Blowwww meeee

Nick Li's picture

kickstarter?

Philipp Dollst's picture

Signed by Adobe in 3, 2...

Igor Butskhrikidze's picture

looks cool

Andrew Sible's picture

So perhaps in a few years this could bring reality to those CSI scenes where the protagonist investigator asks the super-tech security personell to zoom in to 500% and enhance four times from an ordinary security camera hidden in a 10" high ceiling vent to reveal something the size of a dime in crystal-clarity.

"zoom in on grid three, section two, ok magnify -- magnify -- once more -- THERE!! HIS TOOTH STAIN MATCHES THE SAMPLE IMAGE WE HAVE, GOT OUR PERP."

KO KO's picture

HAHAAAAAAAAA

Spy Black's picture

Blade Runner beat them to that in '82...

Andrew Sible's picture

Well yeah there are quite a few instances of that sort of thing, lol.

Timariuveo's picture

de-convolution isnt very new, the problem has always been that without knowing HOW the lens distorts the image, there is no way of figuring out how to correct it.
Since they know the lens, and tested it they know exactly how the lens distorts the images and thus they can correct it.
So, it only works with well tested lenses, even 2 same lenses have different distortion patterns.

njnewsreader's picture

Exactly. De-convolution has also been used to de-blur images. Even if de-convolution is successful, noise can be amplified, and artifacts can be introduced. One can make the image look better but one cannot create the information that is not there.

Spy Black's picture

Well, the information isn't so much not there, as it is not there in a form we can readily process. Back in the early 80s there was an article in I believe Science News that showed the (then) state of the art in satellite imaging. One interesting process (done by Kodak, no less) involved showing an example of a barn up a rocky road with trees around it a sky with puffy clouds, in perfect focus. The next example was the same scene shot completely out of focus. The final scene was a shot reconstructed from the OOF shot, where you could clearly see the everything in the scene. While you could see the data loss compared to the in-focus shot, that fact that you could see anything at all was frankly quite stunning. This was back in the early 80s, mind you.

Bert McLendon's picture

I was initially going to say "Don't sell to Adobe" because I usually think innovation comes when there is pressure to make money to eat, but then I realized that I rent my software now and this would be a really nice "update" one day. I wonder how sharp it could make an already sharp lens? muaahahahahaha!

Daniel Zaleski's picture

Did you check DxO Labs implementation of something very similar in the DxO Optics Pro, the tool is called "DxO Lens Softness". Their deconvolution algorithm is backed by the lens measurements, i.e. you have lens-body measured data and based on these data DOP process the raw data. Works like a charm and really boosts optical quality.

Mike Diblicek's picture

Great results, early days but a game changer for me.

greg tennyson's picture

It'll be a race between Adobe and GoPro to pick these guys up. That software would be very valuable to companies building action sports cameras, where size and weight are important factors in the construction of the camera system.

Keith Pun's picture

I wonder how much Canon and Nikon would pay to not have this software hit the market?

Mark Alameel's picture

Sadly, that is exactly what I was thinking.

Solutions like this could kill the lens market. However, big lens is not about only sharpness but also DOF, bokeh quality, distortion, build quality, etc. That said, there are plenty of quality fast lens that are not thousands of dollars and these "lower" end brands would instantly be bigger competitors. I doubt that we will see this in-camera anytime soon but we can hope for a post effect...

Mr Blah's picture

I don't agree.

The market would change, but on big professional jobs, client will still want to see L series lenses and top of the range optics...

Many pro photogs didn't get gigs because the client thought he couldn't pull it off with non-top of the line gear...

Mark Alameel's picture

Yes, the top of the top will always have the best of the best... however, they are not indicative of the market or a sign of trends in photography. The market will change inspite of them. Then they will also have to change because the cost of that superiority doesn't give better results when it is a commercial job.

It has happened time and time again. here used to be a time when medium format film was the only option. Now our current President is captured with a Canon 5D markII (with the 24-105L lens). Like film, lens will follow. It's the next big thing to crash in prices in photography. Strobes are already much cheaper and way more powerful than when I started. Perceptions will change.

Most photographers are hired by their portfolio, not their equipment lists.

There'll always be stupid clients and their stories will always gain traction but take those with a grain of salt.

P.S. Until then, many would opt for one L lens to get past the stupid client test. If the job was really a big pro job, then the investment would easily be warranted.

johnbp123's picture

Exactly! Well said.

Grey Chen Junyang's picture

They might buy the tech and into the camera itself and making it work only with their latest lenses. selling new cheap to produce, lighter, faster lenses and killing the competition.

gilbycoyote's picture

I rather see this implemented right into fixed lens cameras (X100 and the likes)

Mark Keck's picture

The comparison I would have liked to see is the same pic shot with the same camera and the best possible lens. How does that look?

Spy Black's picture

Cellphones...

Trenton Meador's picture

I want this now. I wouldn't mind paying quite a bit for this software (of course polished software or plugin).