HDC: A subroutine library for image processing using Burt’s hierarchical discrete correlation

Abstract

This paper describes a library of subroutines that efficiently implement Burt’s (1981) hierarchical discrete correlation (HDC). The subroutines are written in Professional FORTRAN for IBM PCs or compatibles. Subroutines are included for both even and odd forms of the HDC, as well as for the corresponding algorithms for pyramid processing. Odd width kernels of up to 7×7 pixels and even width kernels of up to 6×8 pixels can be programmed easily with these subroutines. Thus, this library contains the forms of HDC that are most widely used in image processing and visual modeling. I show the efficiency of this implementation by reporting the execution time of each subroutine for several image sizes. I also present a method that simulates rotation of the kernel in order to allow applying kernels without circular symmetry at the desired orientation.