FOURIER DOMAIN ANTIALIASING RECOVERY
By João Paulo Navarro Barbosa and Leonardo Sacht
Federal University of Santa Catarina
ABSTRACT
Non-linear image operations generate undesirable artifacts, mainly jagged edges. Even if the input image does not present such artifacts, the operations - usually perfomed on the discrete domain - introduce them.
Edge reconstruction in Computer Graphics is approached in two different ways: on the spatial domain and on the frequency domain.
We propose to develop a method that post-processes an image resulting from a non-linear operation on the Fourier domain using spectral properties of the input image and filtering, since the input image contains information about the most appropriate smoothness of the edges. This Fourier domain processing is done on the GPU.
brief description
As shown in the diagram below, the method consists of obtaining the high frequencies of the input image and the low frequencies of the processed image, so that when initializing some parameters the combination of frequencies creates the antialiased processed image.
Some Results
The images below are some results of applying the method to different frequency domain filtering values. More details are available in the paper and images are available for download on the next topic.
Implementation developed on Matlab®(R2016b) with package GUIDE
[1] GPU Processing: DOWNLOAD
[2] CPU Processing: DOWNLOAD
Paper
[1] J3M (in portuguese) : VIEW
Reference files
[1] Test images: DOWNLOAD
How to use the interface(2 minutes)
Contact Us
[1] João Paulo N. B.: jp_nb95@hotmail.com
[2] Leonardo K. Sacht: leonardo.sacht@ufsc.br