Robust Click-Point Linking for Longitudinal Follow-Up Studies

This paper proposes a novel framework for robust click-point linking: efficient localized registration that allows users to interactively prescribe where the accuracy has to be high. Given a user-specified point in one domain, it estimates a single point-wise correspondence between a data domain pair. In order to link visually dissimilar local regions, we propose a new strategy that robustly establishes such a correspondence using only geometrical relations without comparing the local appearances. The solution is formulated as a maximum likelihood (ML) estimation of a spatial likelihood model without an explicit parameter estimation. The likelihood is modeled by a Gaussian mixture whose component describes geometric context of the click-point relative to pre-computed scale-invariant salient-region features. The local ML estimation was efficiently achieved by using variable-bandwidth mean shift. Two transformation classes of pure translation and scaling/translation are considered in th...

• Type: Conference
• Year: 2006
• Where: MIAR
• Authors: Kazunori Okada, Xiaolei Huang, Xiang Zhou, Arun Krishnan
• Keywords: Explicit Parameter Estimation, Local Ml Estimation, MIAR 2006, Spatial Likelihood Model