Similarity Assessment based on Integration of D2 Shape Distribution and Negative Feature Decomposition for 3D Mechanical Parts

This paper proposes a novel measurement scheme of 3D shape similarity that integrates D2 Shape Descriptor and Negative Feature Decomposition (NFD). Using NFD, the scheme firstly converts a 3D mechanical part into a tree structure of geometrical primitives decomposed from the part model, namely Negative Feature Tree (NFT). The D2 shape descriptions of these primitives are then produced for further similarity assessments. We assess the shape similarity on a level-by-level basis between the NFTs of a query part and a candidate part. The weighted sum of the similarity values computed on each level is then used as a measure of the overall similarity between the two parts. Our approach combines the simplicity of D2 shape description while overcoming its insensitivity to negative features with NFD. It performs more consistently than the method of Convex Hull Difference (CHD). A comparison with the assessment results using D2 and CHD demonstrates the effectiveness of the new scheme.