Parametrically deformable contour models for image analysis
Lawrence H. Staib · 1991
A practical system for boundary finding of natural objects in images has been developed. It is based on a new general probabilistic method of boundary finding that allows the incorporation of prior information about the global shape of the target object. Determining the boundaries of objects, and thereby their shape and location, is an important task in computer vision. Segmentation using boundary finding is enhanced both by considering the boundary as a whole and by using model-based global shape information. Previous boundary finding methods have either not used global shape or have designed individual shape models specific to particular shapes. Imperfect image data can be augmented by exploiting the extrinsic information that a model provides. Flexible constraints in the form of a probabilistic deformable model are applied to the problem of segmenting natural objects whose diversity and irregularity of shape makes them poorly represented in terms of fixed features or form. The objects being considered are expected, however, to have a tendency toward some average shape. The parametric model is based on the elliptic Fourier decomposition of the boundary. This is augmented with probability distributions defined on the parameters, which bias the model to a particular overall shape while allowing for deformations. Boundary finding is formulated as an optimization problem using a maximum a posteriori objective function. The best match is found between the boundary, as defined by the parameter vector, and a measure of image boundary strength derived from the image, as biased by the shape prior probability. A computer implementation was constructed and applied to object delineation problems from a variety of two-dimensional images. Results of the method applied to real and synthetic images are presented. Extensions of this method to three dimensions and temporal sequences are outlined.