Quality measures for lightfield interpolation algorithms

3D displays such as autostereoscopic displays have the potential to reproduce 3D scenes more realistically than traditional stereoscopic displays but at the cost of higher bandwidth. Because bandwidth is limited, the 3D scene is calculated and transmitted at a lower sampling rate and reconstructed using interpolation algorithms. One difficulty is quantifying the amount that bandwidth can be reduced for a given reduction in display quality. This paper describes the use of the three quality measures applied to this purpose: the mean-squared error (MSE), Wang and Bovik's Structural Similarity Index (SSIM Index) and a measure based on the optical flow between composite images, the Optical Flow Measure (OFM).

A program was written to capture synthetic 3D data stored as a lightfield at different sampling rates. These were reconstructed using an interpolating low pass filter, a 3D warping algorithm, and the Lumigraph reconstruction algorithm. Of these, the Lumigraph and 3D warp performed the best under the given quality measures, with about 30 times better OFM than the interpolating low pass filter, at an expense of increased computational time.