Feature retrieval is of great importance in shape modelling, in terms of supporting design reuse by obtaining reusable geometric entities. However, conventional techniques for feature retrieval are generally limited to the extraction of feature lines, curve segments, or surfaces, and the feature distortion imposed by feature interaction remains unconsidered. This paper investigates approaches for freeform feature retrieval by means of signal processing techniques. By treating features or regions of interest as surface signals, we employ digital filters to separate the feature signal from that of the domain surface, retrieving the “pure” feature from an existing shape model. Strategies for different model types are elaborated, for instance, the exact feature retrieval method designed for shape models with explicit data structure, such as B-Rep, or other accessible representations; and the signal filtering method for models with structured or unstructured data sets, such as that in mesh or point cloud models. Specifically, in the signal filtering method feature retrieval is implemented by the convolving operator in the frequency domain. By transforming the problem of shape decomposition from geometric extraction in the spatial domain to computation in the frequency domain, the proposed methods not only brings in significant computational efficiency, but also reduces the complexity of problem solving for feature retrieval. Provided examples show that the proposed approaches can achieve satisfactory results for simple geometries, whereas for sophisticated shapes guidelines for the design of dedicated filters are elaborated.
Application of Pattern Filtering Method in Cross-hole Tomography Signal Processing