Image Enhancement using Calibrated Lens Simulations
YiChang Shih1,2*, Brian Guenter2, Neel Joshi2
1MIT CSAIL, 2Microsoft Research
* This work was performed while the first author was an intern at Microsoft Research.
We use the pinhole array setup in (a), lens prescriptions and simulations in (b) to calibrate highly accurate lens prescriptions. With the calibrated lens prescription, we can simulate accurate point spread functions (psf) at any depth to enhance the image, as shown in (c). In (b) we show 3 lens prescriptions. In (c) we simultaneously remove chromatic abberations and spherical abberations of the two inputs taken by the 3 lenses. The simulated lens psf are shown. Please see the paper and supplementary materials for results.

YiChang Shih, Brian Guenter, Neel Joshi, Image Enhancement using Calibrated Lens Simulations. Proc. 12th European Conference on Computer Vision, ECCV 2012
[paper pdf ] [slide pptx] [slide pdf] [poster PDF ] [ supplementary material web site ]
All lenses have optical aberrations which reduce image sharpness. These aberrations can be reduced by deconvolving an image using the lens point spread function (PSF). However, fully measuring a PSF is laborious and prohibitive. Alternatively, one can simulate the PSF if the lens model is known. However, due to manufacturing tolerances lenses differ subtly from their models, so often a simulated PSF is a poor match to measured data. We present an algorithm that uses a PSF measurement at a single depth to calibrate the nominal lens model to the measured PSF. The fitted model can then be used to compute the PSF for any desired setting of lens parameters for any scene depth, without additional measurements or calibration. The fitted model gives deconvolution results comparable to measurement but is much more compact and require hundreds of times fewer calibration images.
None yet.
We would like to thank Edmund Optics for providing their lens prescriptions. Thank Fredo Durand for helpful discussion.