A Fresh Look at Generalized Sampling:
Supplementary Material

This package contains source code and data files associated with the technical paper:

Source code

The file sample_code.cpp contains the C++11 code presented in the appendix of the paper.

It demonstrates the following set of functionalities:

• Fast inverse discrete convolution
• Safe indexing with reflection at boundaries
• Box kernel
• Hat kernel
• Mitchell-Netravali kernel
• Catmull-Rom kernel
• Cardinal Cubic B-spline, a generalized kernel
• Cardinal Cubic O-MOMS, a generalized kernel
• Simple, intuitive upsampling
• Simple, intuitive downsampling
• Fast, incremental upsampling
• Fast, incremental downsampling

Video results

We compare the reconstruction quality of various filter kernels in a number of different scenarios:

• The videos in chirp-120-rotations show the effect of 120 successive image resampling operations under small rotations. (The image was rotated by 3 degrees about its center in each resampling.)
• The videos in chirp-120-translations show the effect of 120 successive image resampling operations under small translations.
• The videos in chirp-animation show the effect of translating an input image and then upscaling it. The resulting animations help reveal errors in phase (spatial distortions) and amplitude (contrast variations).
• The images in checker-400spu show the results of prefiltering a procedural checkerboard scene viewed under perspective projection. The results were obtained using 400 samples per pixel.
• The videos in checker-animation-400spu show videos of the same prefiltered scene as the viewpoint is animated. This animation helps reveal the temporal stability of the prefiltering.
• The images in zonal-400spu show the results of prefiltering the standard zonal plate. The results were obtained using 400 samples per pixel.
• The videos in zonal-animation-400spu show videos of the same prefiltered scene as the viewpoint is animated. This animation helps reveal the temporal stability of the prefiltering.