Harish Palani (CS 194-26)
Light Field Photography with a Hand-held Plenoptic Camera by Ng et al. demonstrated that complex effects like depth refocusing and aperture adjustment can be achieved with any set of images which capture a scene orthogonally to the optical axis. In this assignment, I recreated those effects with lightfield images from the Stanford chess dataset.
To simulate depth refocusing, I took advantage of the fact that the positions of distant objects don't vary significantly as those of nearby objects with respect to the camera's plane of motion. By controlling the amount by which I shift the given images in the grid and subsequently averaging them, I was able to simulate the intended effect to generate images which appear focused at various different depths. The results for the chess dataset are shown in the animated GIF below, with intervals of 0.1 used to scale the shifts within an empirically selected range extending from -3.0 to 0.3.
In order to simulate aperture adjustments, I selected various subsets of the given images which were captured within some set radius of a fixed point. By averaging a progressively larger subset of images, I was able to simulate an increasing aperture, with results shown below on the chess dataset for intervals of size 0.25 within an empirically selected range extending from 0 to 8.