| Summary: | 碩士 === 國立成功大學 === 資訊工程學系碩博士班 === 100 === Figure-ground perception refers to a perceptual experience of assigning a contour to one of the two adjacent regions and that region, the figure, has a definite shape while the other region, the ground, appears shapeless and continue behind the figure. Visual cues that associate with figure assignment process including low-level geometric and high-level semantic properties have been extensively studies in computer vision and psychology. Ambiguous figure-ground perception is a special phenomena by which adjacent regions in a image compete with each other in terms of visual cues such that viewers could switch the perceived figure among regions in any moment. In this thesis, we present a technique to generate images with ambiguous figure-ground perception. Our algorithm is designed so that two input regions manually labeled as figure and ground are registered locally along a shared contour. Then two geometric properties, namely convexity and area, are adjusted to balance the competition in between figure and ground regions. We demonstrate the effectiveness of our technique using a large number of visually promising ambiguous figure-ground images.
|
|---|
