BoSCC: Bag of Spatial Context Correlations for Spatially Enhanced 3D Shape Representation

Area of Science:

• Computer Vision
• Geometric Modeling
• Machine Learning

Background:

• Traditional Bag of Words (BoW) methods struggle with 3D shape representation due to arbitrary mesh resolution, irregular vertex topology, orientation ambiguity, and transformation invariance.
• Existing methods often rely on global perspectives, limiting their compactness and discriminative power for complex 3D shapes.

Purpose of the Study:

• To introduce a novel, spatially enhanced 3D shape representation called bag of spatial context correlations (BoSCC).
• To address key challenges in 3D shape encoding, including resolution, topology, orientation, and transformation invariance.
• To improve the compactness and discriminative capability of 3D shape representations for enhanced classification and retrieval.

Main Methods:

• Developed BoSCC by encoding spatial relationships using an occurrence frequency histogram of spatial context correlation patterns.
• Introduced spatial context correlation to simultaneously capture geometric and spatial information within local 3D regions.
• Modeled vertex spatial context using multi-scale Markov chains to capture intricate spatial relationships.

Main Results:

• BoSCC demonstrates superior compactness and discriminative power compared to global perspective-based methods.
• The proposed method effectively resolves issues related to arbitrary mesh resolution, irregular topology, orientation ambiguity, and shape transformations.
• Experimental results show BoSCC outperforms state-of-the-art methods in global shape retrieval, shape classification, and partial shape retrieval.

Conclusions:

• BoSCC provides a highly discriminative and compact 3D shape representation suitable for various applications.
• The local perspective and spatial context correlation effectively handle complex 3D shape characteristics.
• BoSCC shows significant promise, especially for scenarios with limited data and partial shape retrieval tasks.