A window model for spatial integration in human pattern discrimination

A simple model of human visual pattern discrimination was designed and tested experimentally. The model is based on two assumptions. First, at any glimpse the spatial integration of image information is limited to a window. Second, the observer generates a tailored discriminator for the signals in question using available information. The model is composed of a spatial integration window followed by an ideal discriminator.

The model was tested by comparing its performance with that of human observers in orientation and contrast discrimination. Using a two-alternative, forced-choice method, human orientation and contrast discrimination thresholds were measured for cosine gratings of various areas and spatial frequencies in the presence of two-dimensional spatial noise.

Orientation discrimination thresholds decreased considerably with increasing grating area. Thus, there was clear spatial integration. However, in contrast discrimination, thresholds appeared to decrease only slightly. To make the two tasks comparable, the results also were expressed in terms of efficiency. Human efficiency decreased with grating area in a similar way in the two tasks. This suggests that the factors limiting spatial integration are the same in both tasks. Indeed, the threshold data were explained by the model with the same window size in both tasks with good accuracy. The absolute performance of the model was approximately equal to that of human observers.

The success of the model supports the hypothesis of a spatial integration window. It also supports the idea that human observers use knowledge about the signals to generate an efficient discriminator.