Change-Point Detection for Multi-Way Tensor-Based Frameworks.

Area of Science:

• Data Science
• Computer Vision
• Signal Processing

Background:

• Graph-based methods excel with high-dimensional data but often overlook object features.
• Detecting changes in specific object features (e.g., color) is crucial in many applications.
• Existing methods may not effectively handle feature-specific distribution changes in tensor data.

Purpose of the Study:

• To propose a general graph-based change-point detection method using a multi-way tensor framework.
• To enable the detection of objects with differing features that exhibit distributional changes across tensor slices.
• To enhance detection efficiency by addressing natural disturbances like varying lighting conditions.

Main Methods:

• Developed a general graph-based change-point detection framework utilizing multi-way tensor analysis.
• Incorporated histogram equalization techniques to mitigate the impact of image or video disturbances.
• Applied the proposed methods to simulations and real-world data for validation.

Main Results:

• The proposed multi-way tensor graph-based method effectively detects change-points related to feature distribution shifts.
• The histogram equalization enhancement significantly improves detection efficiency in the presence of natural disturbances.
• Demonstrated superior performance compared to existing methods in simulations and real data analyses.

Conclusions:

• The novel graph-based tensor approach offers a robust solution for feature-specific change-point detection.
• The integration of histogram equalization enhances the practical applicability of change-point detection in real-world scenarios.
• The findings highlight the importance of considering feature variations and data disturbances for accurate change-point detection.