|
Three‐dimensional assets have become a key element in the fields of electronic entertain‐
ment, medicine, and engineering. Unfortunately, 3D models contain large amounts of data that
are not easily suitable for analysis or simple algorithms. Unlike standard datasets with separate
observations and a fixed number of features that can be directly analysed, here the vertices
(data points) are distributed in a highly irregular way. This creates challenges in applying many
algorithms. This study thoroughly examines the usability of the traditional Isolation Forest (IF)
method as a new tool for 3D mesh analysis. Due to the unusual nature of 3D model data, it
was necessary to generate a special multidimensional feature vector (FV) for each vertex.
The FV captures information about the local surface curvature around a vertex. As shown
by experimental results, the IF analysis can identify geometric details, dense and complex
regions, strong bends, and folds in the mesh. Vertices in these areas are classified as anomalies
by IF. Several scenarios and models were analyzed, including different neighbourhood sizes
around a vertex, meshes with tens of thousands of vertices, and low‐poly models. The results
reveal significant steganographic potential, which led the authors to apply these findings
to 3D mesh watermarking as a practical example. Using IF, a new steganographic method was
developed that offers higher transparency, by hiding data in areas of complex geometry. The
study demonstrates the high potential of Isolation Forest for analysing and watermarking 3D
models, marking an important step towards wider use of Isolation Forest in this field.
|
