A Constraint of Invariance in Integral Features across Frequency Bands Produces Illusory Percepts

Much of object perception requires the visual system to integrate image information for image features that are separated in space or in time. Thus, in a system capable of working with real-world images, there must exist detectors specialized to respond to illusory percepts. However, to discover a scheme that can account for tasks such as long-range gure completion (and thus illusory gures), phase-induced boundary detection, and grouping of features poses several problems in computational vision. First, we must de ne what we mean by illusory percept. In this paper, the notion of illusory percept is used to describe structures that a constraint of invariance in integral features across frequency bands produces. Second, there is the problem of the initially unknown recognizer that enables illusory gures and contours to be segregated. To address this problem, we have developed a computational model for measuring the degree of alignment of integral features across frequency bands. Experimental results are provided to indicate the performance of this model on several examples: (i) Kanizsa-type subjective gures; (ii) phase-induced subjective contours; (iii) grouping by similarity; (iv) texture segregation when there are neither global features that allows the region to be segregated from the background nor regions where there exists feature contrast; and (v) the Fraser spiral.