Optimum flux-detection in the absence of a priori knowledge about the signal

Detection systems based on photon counting have to discriminate between two types of fluctuations in the photon count: those resulting from statistical fluctuations (=noise) and those caused by changes in the radiance set by the source (=signal). In contrast with earlier studies on ways of discriminating noise from signal changes, no specific assumptions are made about the source. An optimal discrimination-method has been developed for a detector that has no prior information about the mean of the Poisson distribution that describes its input signal. Because the detector has no prior information at its disposal it has to assume an a priori probability for the mean in a unique and objective way and it has to estimate the actual mean using Bayes rule of inference. This new discrimination-method is discussed in the context of signal processing in the visual system, but is generally applicable in all systems where photon-noise is important.