Problems with modeling detection and identification of signals in noise
Stanley A. Klein, Dennis Michael Levi · Journal of Vision · 2010
A powerful method for peeking into the black box of perception is to measure thresholds for detecting and identifying patterns in noise. A typical experiment varies two parameters: the rms strength of the external noise (Next) and the criterion at which threshold is measured (d'). The Lu/Dosher Perceptual Template Model (PTM) and the Barlow/Pelli Equivalent Noise Model (ENM) claim that the TvN (threshold vs. noise) curve is well fit by a separable function of the form: Th(Next,d')= CRF(d') (1+(Next /Neq)^p)^(1/p) where the contrast response function behaves as CRF(d')∼d'^(1/t) for d'Neq we found t∼1.5±0.3. This finding that the CRF shape depends on Next means that all separable models are inadequate. We also found that Neq was very close to the noise detection threshold.