Shape Guided Object Segmentation

摘要: We construct a Bayesian model that integrates topdown with bottom-up criteria, capitalizing on their relative merits to obtain figure-ground segmentation is shape-specific and texture invariant. A hierarchy of segments in multiple scales used prior all possible segmentations the image. This by our top-down part query detect object parts image using stored shape templates. The detected are integrated produce global approximation for object’s shape, which then an inference algorithm final segmentation. Experiments large sample horse runner images demonstrate strong despite high background variability. robust changes appearance since matching component depends criteria alone. may be useful additional visual tasks requiring labeling, such as scene objects.