Ordered Subspace Clustering With Block-Diagonal Priors.

Area of Science:

• Data Science
• Machine Learning
• Computer Vision

Background:

• Sequential data analysis is crucial in many applications.
• Existing clustering methods often fail to leverage inherent order information in sequential data.
• Subspace clustering for sequential data remains a challenging research area.

Purpose of the Study:

• To propose a novel subspace clustering method for sequential data that effectively utilizes order information.
• To introduce an efficient algorithm for solving the proposed complex optimization model.
• To enhance clustering accuracy by incorporating a block-diagonal prior.

Main Methods:

• Developed an ordered sparse clustering with block-diagonal prior (BD-OSC) model.
• Employed a quadratic normalizer for data sparse representation to capture coefficient correlations.
• Integrated a block-diagonal prior into the spectral clustering affinity matrix.
• Designed an efficient algorithm to solve the BD-OSC optimization problem.

Main Results:

• The BD-OSC method demonstrated superior performance compared to state-of-the-art subspace clustering techniques.
• Experiments on diverse datasets including synthetic data, human faces, video clips, motion tracks, and 3D face sequences validated the method's effectiveness.
• The proposed method successfully utilized sequential order information for improved clustering.

Conclusions:

• The proposed BD-OSC method offers a significant advancement in subspace clustering for sequential data.
• The integration of a quadratic normalizer and block-diagonal prior effectively models data correlations and improves accuracy.
• BD-OSC provides a robust and efficient solution for analyzing complex sequential datasets.