電子計算機による画像処理におけるフィルタリングのための高速アルゴリズム

概要

論文の詳細を見る
The computation time of performing two-dimensional digital filtering by the computer is proportional to the size of the local region used in the filtering. For example, if a rectangular local region with the size K×L is adopted, the average computation time per sample point in the output image is proporitional to K・L. In this paper, first, three kinds of fast algorithms are proposed for a uniform weight linear filter and a local maximum filter. The computation times of these algorithms are proportional to K+L in the worst case and are approximately independent of K and L in most cases,Next, these algorithms are implemented as FORTRAN subroutines, and their computation times are estimated experimentally. It is known that these algorithms are much faster than other ones thus far developed. Finally, some comments are presented on deriving fast algorithms of filterings by using resolution of filterings, recursive expressions, and FFT.
1976-03-15