Improved Mesh Segmentation with Perception-Aware Cuts

Abstract

High quality mesh segmentation depends on high quality cuts. Unfortunately, the cuts produced by existing methods are not very satisfactory, since their global measurements tend to ignore effects of local features, while their local measurements would enlarge the influences from facet details by error accumulation. We observe that the cuts preferred to by human beings are much more dependent on the overall characteristics of local regions, a kind of intermediate-level features, especially in concave regions. Thus, we present a construct to enhance representation of overall characteristics in concave regions for improving cut initialization in concave regions, and design novel energy functions, mainly by intermediate-level features, for extending cutting lines to be enclosed. Then, based on the obtained closed cutting lines, we perform meaningful mesh segmentation in a bottom-up manner according to application requirements. In comparison with state-of-the-art methods, we can have cuts produced more preferred to by human beings, as shown by the experimental results on a benchmark.