Speeding Up the Computation of Cross-Correlation Using the FHT and FFT by Page-locking

Abstract

Cross correlation is widely used for image matching. The fast Fourier transform (FFT) and the fast Hartley transform (FHT) are used to speed up cross correlation computation. The Hartley transform is specifically designed for real-valued data. FHT is especially useful on small-memory machines as it is its own inverse. However, cross correlation based on FHT suffers heavily as compared to FFT, in paged operating systems, as it requires more page frames to be present in the physical memory at any time. Thus chances for page faults in a high memory demand environment increase, thereby worsening the computation times. In this paper we use the concept of page locking to minimize the number of page faults in both FFT and FHT based cross correlation computation. We also combine several consecutive steps in FHT based method to further reduce the number of page faults by half.