Noisy Template Matching: A Model for Detection and Discrimination

The detection and discrimination of simple patterns occupies a central place in visual psychophysics. A wide variety of phenomena have been observed in this paradigm, such as: Weber's law; masking (simultaneous, forward, and backward); masking by noise; spatial frequency tuning; orientation tuning; and area summation. We suggest that many of these phenomena can be explained by a simple model which we call ‘noisy template matching’.

In this model, the encoded stimulus is matched to a memorised template. Both stimulus and template are corrupted by additive noise. The template matching operation yields a decision variable, to which more noise is added. This model is very simple, but it has many interesting consequences. It provides qualitative explanations for many of the phenomena mentioned above, and with additional (but we think reasonable) assumptions about lens blur, contrast nonlinearity (Whittle, 1986 Vision Research 26 1677 – 1691), uncertainty (Pelli, 1985 Journal of the Optical Society of America 2 1508 – 1532), and suboptimal templates, the model also provides good quantitative accounts of these phenomena.