Contour shape matching by a constraint satisfaction network

The contour shape-matching task is formulated as a constraint satisfaction problem and solved using a Hopfield net. Constraints that are intrinsic to the shape-matching task are coded in the network interconnection weights. Direct evidence obtained from the observed data, as gauged by a local shape measure, are provided as the external input to the network. The local shape measure used is based on an affine transformation model, which is a general class of mappings that includes the rigid body motion. Empirical results demonstrate the performance of this network when applied to match contours that have undergone rotation, scale change, translation, or combinations of these transformations. The network is shown to be capable of handling noisy data, as well as a reasonable amount of deformation resulting from object rotation in the three-dimensional space. The affine transformation can be generalized to match entire contour segments; hence the network is not confined to matching point patterns