作者: Stephen T. Barnard , William B. Thompson
DOI: 10.1109/TPAMI.1980.4767032
关键词: Parallax 、 Computer vision 、 Artificial intelligence 、 Iterative reconstruction 、 Algorithm design 、 Matching (statistics) 、 Point (geometry) 、 Statistical classification 、 Similarity (geometry) 、 Binocular disparity 、 Mathematics
摘要: An algorithm for matching images of real world scenes is presented. The a specification the geometrical disparity between and may be used to partially reconstruct three-dimensional structure scene. Sets candidate points are selected independently in each image. These locations small, distinct features which likely detectable both images. initial network possible matches two sets candidates constructed. Each match specifies point reference estimate probability made, based on similarity subimages surrounding points. estimates iteratively improved by relaxation labeling technique making use local continuity property that consequence surfaces. effective binocular parallax, motion object motion. It quickly converges good disparity, reflect spatial organization