Abstract
The visual world is a complex jumble of lines and edges representing objects and their backgrounds. The visual system groups these signals to form the boundaries of contours so that objects can be segmented from each other to recognize and interact with them appropriately. Integration of information around contours has been demonstrated with low radial frequency (RF) patterns but cannot be demonstrated for all patterns. Biederman has suggested that matched concavities can trigger segmentation of a contour. This project uses integration as a tool to examine when matched concavities lead to the segmentation of complex contours into elemental parts. This study uses radial frequency patterns, deformed circular shapes, which drive global contour integration, as elements, extending previous work to look at more complex shapes. Three psychophysical shape discrimination experiments were conducted, investigating contour integration within patterns, between two separate patterns, and in patterns that are overlapped to vary the depth of contour concavities to either form a single object or an interpretation of two overlapping objects. Two interval forced-choice shape discrimination methods were used to intensively investigate rules driving performance in six observers. The results show that information interpreted to be on an unsegmented complex contour did not reflect integration, while information interpreted to be on continuous contours perceptually segmented into elemental RF patterns was integrated within an element but the elements were separately processed and performance was precisely predicted by assuming probability summation between two separately integrated patterns. Overall the research suggests that while contour integration is seen when observers are detecting elementary shapes, more complex patterns, even if produced by combining those elementary shapes, do not always yield results reflecting such contour integration. Segmentation at matched deep concavities does allow the integration process on the constituent elements but without the segmentation integration is not observed.
Meeting abstract presented at VSS 2018