Modeling, Approximation, and Estimation of Spatially-Varying Blur in Photographic Systems

Several image restoration and analysis approaches, such as deconvolution, super-resolution, and depth-from-defocus, require an accurate representation of the photographic blurring process. Inaccurate representation of the blur which resulted in a particular observation results in inferior restoration of the observation, and potentially an inability to perceive the latent content present the original unblurred scene.

In this dissertation, we propose three novel contributions towards the more accurate representation of photographic blur. We propose a model which constrains the spatial variation of blur across the image plane, and is strongly motivated by the underlying optics model of the camera. We also propose a quantitative measure of blur estimation accuracy, and of measuring the inaccuracy introduced by using an approximation to the true blur in deblurring.

Finally, we introduce an improved method of non-blind blur estimation, and demonstrate the accuracy benefit of using the proposed active target estimation instead of the traditional passive target estimation.