Abstract — This paper presents a new algorithm that finds the best (partial) match between a parametric model curve and a data segment from acquisition. Dynamic programming (DP) is employed for efficiency of computation and combined with the handling of noisy data for robustness. The optimal transformation and scaling in an entry of the DP table are determined from maximizing a probability function, which is solved with an extension to the ICP algorithm [1]. To control the matching accuracy, we have introduced a statistical characterization of dissimilarity. The algorithm has been demonstrated over synthetic and range data. The experiment shows that it adjusts well to the noise distribution of data and performs effectively over curves of complex shapes. An application to model-based 3-D object recognition is also presented. I.