Abstract

Signal representation by its points of crossing of a reference signal is considered with emphasis on applications to images. Some results that are elaborated for the representation of one-dimensional signals by their sine-wave crossings are extended and applied to images. The issue of the stability of the reconstruction algorithms is examined through derivation of bounds on the mean-square error caused by inaccuracies in the determination of crossing locations and by the effect of out-of-band energy. These bounds highlight the dependence of the reconstruction error on major parameters of the underlying signal and the reference sine Wave. Computational results on images reconstructed from their sine-wave crossings support the theoretical results and indicate that the bounds are tight. The representation of signals by their crossings of a reference signal is further examined in the context of a general framework of unique representation wherein the signal is modeled as a sample path of an ergodic band-limited random process. The idea of a scheme for signal representation that is almost stable in a probabilistic sense is introduced and discussed.

© 1987 Optical Society of America

Reconstruction of multidimensional signals from zero crossings

Susan R. Curtis and Alan V. Oppenheim
J. Opt. Soc. Am. A 4(1) 221-231 (1987)

Fingerprints theorems for zero crossings

A. L. Yuille and T. Poggio
J. Opt. Soc. Am. A 2(5) 683-692 (1985)

Pattern and motion vision without Laplacian zero crossings

John G. Daugman
J. Opt. Soc. Am. A 5(7) 1142-1148 (1988)

Efficiency of a model human image code

Andrew B. Watson
J. Opt. Soc. Am. A 4(12) 2401-2417 (1987)

Segmentation of range images using sine wave coding and Fourier transformation

M. Rioux, P. Boulanger, and T. Kasvand
Appl. Opt. 26(2) 287-293 (1987)