Instead of using convolution to produce the imagery, I think it should be produced using fractional Fourier transforms. IIRC fractional Fourier transforms are mathematically equivalent to Fraunhofer and Fresnel diffraction integrals.
Although, the convolutions look good.
I'm also not super up-to-date on my Fourier optics but I had to research again a bit, as I had something in the back of my brain telling me: Hey, there was something with simple convolutions in Fourier optics and I know spatial convolution with a response function is used extensively to gauge the resultion of your system's image due to diffraction in the system.
Within the Fresnel approximation, the output field can either be formed in a frequency-domain approach with a spectrum of plane waves or as a spatial superposition of paraboloid waves (in which the transfer function is inverse-fourier transformed and then convolved with the input field). And the latter approach is simply a convolution of the input field (in position space) with an impulse-response function of the linear (and shift-invariant) optical system between. Shift invariance (i.e. the response function itself isn't position dependent) is an okay approximation for the central FOV of the human eye.
So what he's doing, the formalism, is kinda correct within the Fresnel regime, only he uses an approximation for the impulse-response function itself. This impulse-response function is called the PSF or Point-Spread Function in imaging optics design, defined as the image (including diffraction effects of course) of a point source. His approximation as a disk is okay-ish, qualitatively (google "PSF of human eye defocus") but I didn't check the numbers for the size.
