on modelling, extraction, detection and classification of deformable contours from noisy images

摘要

We present an integrated approach fn modelling, extracting, detecting and classifying deformable contours directly from noisy images, based on the generalized active contour models (g-snakes) [ 1 ]. Our contour representation for an arbitrary shape is stable and regenerative, as well as invariant and unique under affine motions. We combine this shape model with Markov random fields to yield prior distribution that exerts influence over the arbitrary shape while allowing for deformation. Using our formulation, low level visual tasks of shape modelling and extraction can be readily integrated with high level detection and classification.