Efficient subpixel image registration algorithms

Abstract

Three new algorithms for 2D translation image registration to within a small fraction of a pixel that use nonlinear optimization and matrix-multiply discrete Fourier transforms are compared. These algorithms can achieve registration with an accuracy equivalent to that of the conventional fast Fourier transform upsampling approach in a small fraction of the computation time and with greatly reduced memory requirements. Their accuracy and computation time are compared for the purpose of evaluating a translation-invariant error metric.

Optimizing disk registration algorithms for nanobeam electron diffraction strain mapping

“…Therefore, in order to obtain the most accurate strain measurements, subpixel shifts in disk position must be measured. While many subpixel registration algorithms exist [17,18,19,20], several considerations must be taken into account when choosing which to use. Due to the incredibly large size of the datasets obtained, both in file size and number of disk positions to be measured, only the fastest algorithms are considered.…”

“…The second, faster, algorithm involves Fourier upsampling in the frequency domain and is well described in [18,25]. Using the original correlation peak as its center, the method directly computes subpixel IFFT values a fixed range around the brightest pixel with an arbitrary upsampling factor κ.…”

Isocenter verification for linac‐based stereotactic radiation therapy: review of principles and techniques

“…Their work was based on a previously‐developed technique by Guizar‐Sicairos et al ( 67 ) The method was applied seven times at distinct angles and compared to measurements made by: (a) human observer using templates designed as a graphical user interface, and (b) an edge detection and center of mass algorithm. The proposed method was faster than both due to the higher speed of calculations in Fourier domain.…”

Coherent X-ray diffractive imaging on the single-cell-level of microbial samples

“…This can be done by calculating the cross-correlation between adjacent projections [113], which, however, may need manual corrections. Equal results have been obtained more quickly by using a centre of mass approach [270] in combination with the fact that the specimen here is isolated and that its support χ(x, y, z)…”