Spectral 3D mesh segmentation with a novel single segmentation field

We present an automatic mesh segmentation framework, which achieves 3D segmentation in two stages, comprising hierarchical spectral analysis and isolinebased boundary detection. During hierarchical spectral analysis, a novel single segmentation field is defined to capture concavity-aware decompositions of eigenvectors from a concavity-aware Laplacian. Specifically, on the eigenvector hierarchy, a sufficient number of eigenvectors is first adaptively selected and simultaneously partitioned into sub-eigenvectors through spectral clustering. Next, on the sub-eigenvector hierarchy, we evaluate the confidence of identifying a spectral-sensitive mesh boundary for each sub-eigenvector by two joint measures, namely, inner variations and part oscillations. Selection and combination of sub-eigenvectors are thereby formulated as an optimization problem to generate a single segmentation field. During the isoline-based boundary detection, segmentation boundaries are recognized by a divide-merge algorithm and a cut score, which respectively filters and measures desirable isolines directly from the concise single segmentation field. Experimental results on the Princeton Segmentation Benchmark and a number of complex meshes demonstrate the effectiveness of the proposed method, which is comparable to recent state-of-the-art algorithms.