Since the 1970s and 1980s, the standard view of spatial vision has been that contrast detection involves probability summation over multiple short-range filter-elements, typical of those found in the primary visual cortex. However, more recent work has favoured a model involving physiological integration of luminance contrast over substantial parts of the image, concluding that in previous studies, the effects of a square-law transducer followed by linear summation of noise with the signal have combined to masquerade as probability summation. Here we addressed the details of the contrast integration process by comparing it to an ideal observer model, which was progressively adjusted until it fitted our area summation results (1 to 32 cycles square). We argue that observers use spatially extensive templates (with diameters of several cycles) that operate on the luminance contrast image, but only after retinal inhomogeneity, short-range filtering, square-law transduction and the addition of internal noise. Ideal templates that are matched to the image after the effects of the three processing stages above, provide the modeler with a convenient implementation but might imply a greater level of sophistication than is needed to explain the results.
ECVP 2012 Abstracts