Model-based image processing techniques have been proposed as a way to incr
ease the resolution of optical microscopes. Here a model based on the micro
scope's point-spread function is analyzed, and the resolution Limits achiev
ed with a proposed goodness-of-fit criterion are quantified. Several experi
ments were performed to evaluate the possibilities and limitations of this
method: (a) experiments with an ideal (diffraction-limited) microscope, (b)
experiments with simulated dots and a real microscope, and (c) experiments
with real dots acquired with a real microscope. The results show that a th
reefold increase over classical resolution (e.g., Rayleigh) is possible. Th
ese results can be affected by model misspecifications, whereas model corru
ption, as seen in the effect of Poisson noise, seems to be unimportant. Thi
s research can be considered to he preliminary with the final goal being th
e accurate measurement of various cytogenetic properties, such as gene dist
ributions, in labeled preparations. (C) 2000 Optical Society of America. OC
IS codes: 100.6640. 170.2520, 110.4280, 110.0180, 120.4640, 100.2960.