Bayesian Blended Landmark Model for Alignment of Functional Data

Studies involving functional data often require curve registration - namely, the alignment of salient features in the temporal domain - as a preliminary step before applying inferential techniques. This process reduces phase variability, enabling a focus on amplitude variability. In this work, we introduce a Bayesian model for curve alignment and apply it to a biomechanical dataset comprising three groups of patients. The proposed model strikes a balance between flexible smoothing and effective alignment. Additionally, it leverages landmark points as prior information through a heuristic algorithm to further enhance the alignment process.