‘Region-growing’ algorithm for matching of terrain images

摘要

The paper describes and discusses a new algorithm for stereo matching, which has been designed to work well with data from the SPOT satellite1. It is basically an extension of Gruen's adaptive least squares correlation algorithm2.3, so that whole images can be automatically matched, instead of just selected patches. Initial results on the quality of results are presented and discussed, together with a theoretical analysis of the potential speed on both conventional and multiprocessor architectures, and a brief discussion of the running speeds achieved. In addition, a method for assessing the accuracy of the matching without using any external reference data is presented.