For geometric inference from images, what kind of statistical model is necessary?
Kenichi Kanatani · Systems and Computers in Japan · 2004
Abstract In order to promote mutual understanding with researchers in other fields including statistics, this paper investigates the meaning ofstatistical methodsfor geometric inference based on image feature points. We trace back the origin of feature uncertainty to image processing operations and discuss the meaning ofgeometric fitting, geometric model selection, thegeometric AIC, and thegeometric MDL. Then, we discuss the implications of asymptotic analysis in reference tonuisance parameters, theNeyman‐Scott problem, andsemiparametric modelsand point out that application of statistical methods requires careful considerations about the peculiar nature of geometric inference. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 35(6): 1–9, 2004; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/scj.10635