Abstract: PURPOSE: As a fundamental bottleneck in object recognition, crowding is typically considered as excessive integration of nearby items, resulting in impairment in target identification. However, the integration strategy in visual crowding remains unclear. Here we proposed a new integration model in visual crowding, and tested if it can best account for the observation compared to other models.
METHODS: We adopted a magnitude matching paradigm (Baldassi et al., 2006; Gheri & Baldassi, 2008) to probe the internal response of target orientation in clutter. Subjects were asked to report both the direction and the magnitude of a target with surrounding flankers in a crowded display. We varied the number of signal as well as the tilting angle of the signal. In addition, the signal items were placed either along the radial or the tangential axis with respect to the fovea fixation. By constructing the response distributions, we tested which model best accounts for the observation across signal numbers, tilting orientations, and radial-tangential layouts.
RESULTS: 1) In contrast to the prediction of a pure averaging operation, we observed a bimodal response distribution, supporting an involvement of signed-max operation in the late stage. 2) Crowding can be modelled with radially elongated averaging zone, and was better accounted for when the signals were positioned in a congruent spatial layout. 3) An increase of signal number was associated with higher precision in orientation judgement, which can also be predicted by the mixed model.
CONCLUSIONS: Our findings suggest a hybrid strategy in combining crowded signals: averaging in early visual processing stage followed by a signed-max operation in higher-level processing stage.

Key words: visual crowding, integration strategy, psychophysics