In the computer vision community, the usual way of simulating the defocus blur is by convolving the original image with a blur kernel. This kernel is usually chosen to be;
- Gaussian, or
- 2D Bessel kernel (i.e., Airy Disk), or
- A simple disk kernel.
For all of these kernels, the amount of the blur is controlled by the size of the kernel. However, this is not a realistic model of defocus blur. For a more physically meaningful defocus:
- Zernike Polynomials based defocus aberration, and
- Hanser's method described in Phase retrieval for high-numerical-aperture optical systems,
are implemented and can be found in the Notebook. As an example, here is a comparison of the Zernike Polynomial based defocus vs. Disk convolution blur:
The implementations of the two fundamental image deblurring algorithms;
- Wiener Filtering, and
- Richardson Lucy Algorithm
can be found in the Notebook. As an example, here is the results of applying a naive inverse filtering approach vs. Wiener Filtering for both known and unknown Signal-to-Noise ratios in the presence of small additive noise:
