Approach for 3D Cultural Relic Classification Based on a Low-Dimensional Descriptor and Unsupervised Learning

Area of Science:

• Digital Archaeology
• Computer Vision
• Cultural Heritage Management

Background:

• Virtual cultural relic management relies on computer-aided classification.
• Existing methods often lack sufficient category labels or have uneven data distribution.
• This poses challenges for practical applications in digital heritage.

Purpose of the Study:

• To propose a 3D cultural relic classification method robust to data scarcity and imbalance.
• To enable effective virtual management and display of cultural relics.
• To advance digital archaeology and virtual reality applications.

Main Methods:

• Computation of scale-invariant heat kernel signature (Si-HKS) descriptors.
• Transformation of Si-HKS into a low-dimensional feature tensor using Bag-of-Words (BoW) mechanism, creating the SiHKS-BoW descriptor.
• Application of the unsupervised learning algorithm MKDSIF-FCM for classification.

Main Results:

• The proposed SiHKS-BoW descriptor combined with MKDSIF-FCM achieved a classification accuracy of up to 99.41%.
• The method effectively addresses challenges of missing category labels and uneven data distribution.
• Validation was performed on a dataset of 3D models of Tang tri-color Hu terracotta figures.

Conclusions:

• The developed unsupervised learning method provides a practical solution for 3D cultural relic classification.
• This research enhances the application of virtual reality in digital heritage projects.
• It contributes to enriching digital archaeology content and management strategies.