Design and theoretical analysis of a vector field segmentation algorithm

Several objective functions for vector field segmentation are presented. Y. G. Leclerc's (1989) MRF (Markov random field) model is extended by the addition of information-theoretic penalties for regions and distinct means. Standard methods of signal detection and estimation are used to develop a theoretical performance analysis which quantitatively predicts the performance at realistic noise levels. The theoretical performance analysis demonstrates the need for qualitative change from the scalar case; separate penalties for boundary structure and region existence are very beneficial for high d (dimensional). The theoretical analysis also indicates the merit of an objective function before an optimization algorithm has been developed. It also serves as a benchmark for optimization algorithm performance. Theoretical and experimental results agree fairly well.<>