Parallelization of the resampling image filter

Abstract EN

When resembling a digital image by uniform cubic B-splines, an output pixel is computed from a filter applied to 16 neighboring pixels of the original image or more precisely of some auxiliary matrix C.

Matrix C is computed with a computational time proportional to the number of pixels of the image to be resembled ; that time is substantially smaller than the computational time of the filtering part.

In this paper, an adapted Cholesky factorization is presented ; it allows to calculate the matrix C, the resembling filter of the 16 neighboring pixels and a parallel computation of the filter.

This parallel approach reduces the global computational time partly from a better memory management.