Model for the detection of line signals in visual noise

Two experiments are described that investigate the detectability of signals embedded in spatially uncorrelated visual noise. The signals were thin lines extending vertically from the top to the bottom of the noise background and were essentially of two classes. One class consisted of signals whose pixel luminances were spatially uncorrelated, the other of signals whose pixel luminances were sinusoidally modulated along their length. In both classes the space-average luminance of the signals was greater than that of the noise. In the first experiment it was found that the modulated signals were more visible than the unmodulated, even though the space-average luminance of both types was equal. For the modulated signals visibility was proportional to the amplitude of modulation and inversely proportional to the spatial frequency of modulation. Neither an ideal-observer model nor an energy-integrator model was found to give a good account of the results. On the other hand, a model referred to as the maximum filter output model gave an excellent account of the data. The model employed a simple energy-integrating filter approximately 0.8 deg in length and recorded the filter's maximum output for every potential signal, indicating the signal as the line that gave the largest output. In a second experiment, modulated signals that contained added random luminance perturbation were used, and again the model was found to give a good fit to the data. The findings are discussed in the context of known properties of cortical-bar detectors.