Fourier phase estimation is based on the principle
that the phase of the Fourier transform of an image is
a linear function of the displacement. The reference
image of the target (the entity function [11]) and the
observed image are divided pixel by pixel. The phase
of the resultant complex-valued image is a linear
function of pixel coordinates with a slope
corresponding to the position of the target. A least
squares fit to the slope of the phase is used as the
position estimate. As noise is added, the otherwise
erratic performance curve of Fourier phase estimation
is smoothed somewhat. The method does not perform
well until the number of quantization levels gets
quite large, at which point it surpasses the center of
mass method of calculating the centroid.
Centroid estimation is done using the "center of
mass" calculation as was discussed in section 4. This
estimate is straightforward to compute. Explicit
calculations have shown that for the 3 by 3 window
it exhibits a bound of 24 on its reciprocal RMS error
in the absence of quantization. With coarse
quantization, this bound is exceeded to a level of
about 35 (with 14 quantization levels), illustrating
the perversity of quantization effects.
Position decoding is consistently and clearly superior
to the other algorithms both with and without noise
and at all quantization levels except possibly near-
binary. Improvement with the number of
quantization levels is extremely linear (up to an
beyond 100 quantization levels), as predicted by the
optimality criterion of the locales theory. Noise
degrades performance somewhat but linearity is
preserved. The gap between the lineal bound and
optimal position estimation in the absence of noise
can be attributed to the fact that the lineal bound is
based on the limiting assumption that all of the
locales have the same size. It is reasonable to
speculate that the divergence of the two curves can be
derived from knowledge of the distribution of locales
sizes (or perhaps even the converse).
62
SUMMARY
The possibility of acquiring very high quality digital
imagery brings the analysis of quantization effects
into the forefront of digital image metrology. Sub-
pixel position estimation to very high levels of
precision are possible in theory. A foundation of
solid principles is needed for the effective
accummulation of knowledge and experience in high
precision measuring techniques. The theory of locales
is potentially such a unifying basis. Locales are
simply defined and easily generated. They serve to
estimate position as well as the position uncertainty
due to quantization. They lead to a natural and easily
stated definition of optimal position estimation. The
optimal position estimation algorithm called
"position decoding" has been introduced here for the
first time. Preliminary performance simulations have
shown it be an effective tool for digital image
metrology.
REFERENCES
[1] C.A. Berenstein, L.N. Kanal, D. Lavine and
E.C. Olson, "A Geometric Approach to Subpixel
Registration Accuracy", Computer Vision, Graphics,
and Image Processing, Vol. 40, pp.334-360, 1987.
[2] H.A. Beyer, "Some Aspects of the Geometric
Calibration of CCD-Cameras", ISPRS Intercomm.
Conf. on Fast Processing of Photogrammetric Data,
Interlaken, Switzerland, June 2-4, pp.68-81,1987.
[3] J. Dähler, "Problems in Digital Image Acquisition
with CCD Cameras”, ISPRS Intercomm. Conf. on
Fast Processing of Photogrammetric Data, Interlaken,
Switzerland, June 2-4, pp.48-59,1987.
[4] L. Dorst and A.W.M. Smeulders, "Best Linear
Unbiased Estimators for Properties of Digitized
Straight Lines", JEEE Trans. on PAMI, Vol. PAMI-8,
No. 2, pp.276-282, March 1986.
[5] L. Dorst and R.P.W. Duin, "Spirograph Theory:
A Framework for Calculations on Digitized Straight
Lines", JEEE Trans. on PAMI, Vol. PAMI-6, No. 5,
pp.632-639, Sept. 1984.
[14] L. Dorst and A.W.M. Smeulders, "Discrete
Representation of Straight Lines", IEEE
Trans. PAMI, Vol. PAMI-6, No. 4, pp.450-463,
July 1984.
[6] Fairchild, CCD SOLID STATE IMAGING
TECHNOLOGY", Fairchild CCD Imaging,
4001 Miranda Ave., Palo Alto, Ca., 159 pages,
Parat, Munich, 1981.
[7] R.M. Gagliardi and S. Karp, OPTICAL
COMMUNICATIONS, JOHN WILEY & SONS, 1976.