Many image restoration and analysis approaches in the literature rely on an accurate characterization of the linear blur kernel for an image, the point-spread function (PSF). Existing PSF models are either parameterized and spatially-invariant, or spatially-varying and discretely-defined.
In this thesis, we propose a parameterized, spatially-varying PSF model to describe the blur due to lens aberrations and defocus. The model follows from the combination of several geometric camera models, and the Seidel third-order aberration model.
We propose a novel estimation algorithm for computing the parameters of the aberration model from a set of PSF observations, and we demonstrate through simulation that this yields a more reliable set of PSF estimates. In simulated PSF sets with spread measure noise as strong as 10 dB SNR, the proposed model consistently led to PSF estimates with a 5 dB SNR improvement over the observations, and typically a 10 dB SNR improvement.