Ricci flow embedding for rectifying non-Euclidean dissimilarity data
作者：
Highlights：
• Rectify non-Euclidean similarity data by applying Ricci flow to its kernel embedding.
• Euclidean rectification inflates a piecewise manifold on which the data resides.
• Method iteratively flattens the data manifold by increasing its radius of curvature.
• Experiments on shape analysis highlight some of the strengths and weaknesses of the method.
• Proves effective on the Delft Chicken pieces dataset.
摘要
Highlights•Rectify non-Euclidean similarity data by applying Ricci flow to its kernel embedding.•Euclidean rectification inflates a piecewise manifold on which the data resides.•Method iteratively flattens the data manifold by increasing its radius of curvature.•Experiments on shape analysis highlight some of the strengths and weaknesses of the method.•Proves effective on the Delft Chicken pieces dataset.
论文关键词：Non-Euclidean pairwise data,Similarity,Metric,Ricci flow,Embedding
论文评审过程：Received 19 October 2013, Revised 24 March 2014, Accepted 21 April 2014, Available online 5 May 2014.
论文官网地址：https://doi.org/10.1016/j.patcog.2014.04.021
