Researchers produce algorithm to improve image quality
A group of researchers at the University of British Columbia and University of Siegen came with a set of computational photography techniques able to dramatically improve the image quality of pictures taken with DSLR cameras equipped with simple, single element lenses.
Replacing complex camera lens with simpler alternatives
A lens for the modern cameras is a very complex device: as every glass element introduces its own image aberrations and distortions, the manufacturers have to put multiple glass elements, of various shapes and sizes, with the purpose of cancelling-out each others’ aberrations. As a result, a good quality lens can have up to over 20 elements, with direct consequences on its price and weight.
In a paper made for the prestigious SIGGRAPH conference, a group of Canadian and German researchers come with an alternative approach. Instead of the complex lenses of today, they return to the basics and use a very simple and inexpensive lens, as it was used in the old days, and then resort to some clever algorithms to improve the resulted image in post-processing. The algorithms are not magic, but hard math and the core of this paper.
Simple lenses producing better image quality using the algorithms
For the purpose of the research, they used a contemporary digital SLR camera with only custom-made single lens elements such as plano-convex or biconvex lenses, and achromatic doublets. The result was image quality comparable with that of commercial point-and-shoot cameras for apertures around f/4.5, but with degradation in quality for larger apertures of f/2 and beyond. While single lens elements won’t replace a high-end lens, using such techniques may contribute to simplification of lens construction and, therefore, to a decrease in price and weight.
Tested on a commercial lens like Canon 28–105 mm, using this method can still improve the resulted image, so the algorithms can have general purpose use, not being tied exclusively to single lens elements.
The researchers still have a long way ahead to improve the lens
There is further work to be done; currently, the researchers are deconvolving the images with point spread functions (PSFs) calibrated for a single scene depth. This can be improved by performing deconvolution with PSFs for different image depths and wavelengths. Such an optimization can be done knowing the parameters of the optical system. There is a lot of room for improvement and finally we may have simple lenses producing better image quality.
If you are interested in the fine details and the accompanying math, the full paper is available for reading.