Contour integration by the human visual system: Evidence for a local “association field”

The Gestalt law of "good continuation" has been used to describe a variety of phenomena demonstrating the importance of continuity in human perception. In this study, we consider how continuity may be represented by a visual system that filters spatial data using arrays of cells selective for orientation and spatial frequency. Many structures (e.g. fractal contours) show a form of redundancy which is well represented by the continuity of features as they vary across space and frequency. We suggest that it is possible to take advantage of the redundancy in continuous, but non-aligned features by associating the outputs of filters with similar tuning. Five experiments were performed, to determine the rules that govern the perception of continuity. Observers were presented with arrays of oriented, band-pass elements (Gabor patches) in which a subset of the elements was aligned along a "jagged" path. Using a forced-choice procedure, observers were found to be capable of identifying the path within a field of randomly-oriented elements even when the spacing between the elements was considerably larger than the size of any of the individual elements. Furthermore, when the elements were oriented at angles up to +/- 60 deg relative to one another, the path was reliably identified. Alignment of the elements along the path was found to play a large role in the ability to detect the path. Small variations in the alignment or aligning the elements orthogonally (i.e. "side-to-side" as opposed to "end-to-end") significantly reduced the observer's ability to detect the presence of a path. The results are discussed in terms of an "association field" which integrates information across neighboring filters tuned to similar orientations. We suggest that some of the processes involved in texture segregation may have a similar explanation.