Two error-diffusion-based binarization methods for joint transform correlator configurations, which adaptively take into account the effects of input-additive white Gaussian noise, are analyzed. Before binarization, the operations performed upon the joint power spectrum are either truncation and normalization or subtraction of a noise pedestal followed by truncation and normalization. The noise-pedestal value is defined as the measurable estimate of the noise power spectral density. Truncation and normalization are carried out with a spatially constant noise-dependent range limit, based on the statistical properties of the noise, and the noise-pedestal value. All required parameters, dependent on the input-noise level, can be measured from the joint power spectrum distribution and are updated for every new input scene. A computer-simulation comparison of correlation-peak characteristics demonstrates the advantages of the suggested approaches. Optical experiments with compatible results are also presented.