| Abstract: | A general framework for the analysis of sequences of stereo images is presented in this article. The proposed paradigm integrates spatial and temporal information to achieve robustness in establishing stereo correspondence from images that possess a limited number of salient features. Specifically, the recovery of dynamic stereo information, and thus, the reconstruction of a three-dimensional object structure, is accomplished using a feature-based approach. The multiset feature correspondence problem is cast into a four-dimensional problem which is subsequently solved using a greedy-type local search algorithm. No assumption is made about the type of motion. The proposed algorithm may be applied to process image sequences containing both rigid and nonrigid shapes. The efficacy of the proposed approach is tested using both synthetic data and real images of a live frog's ventricle acquired by a stereo light microscope. We also consider the incorporation of domain-specific knowledge to enhance the performance of the cooperative matching scheme. © 1998 John Wiley & Sons, Inc. Int J Imaging Syst Technol, 9, 192–200, 1998 |
