Efficient Multi-View -Means for Image Clustering.

Area of Science:

• Data Science
• Machine Learning
• Clustering Algorithms

Background:

• Real-world data often originates from multiple sources, posing challenges for traditional clustering.
• Existing multi-view K-Means methods struggle with linearly non-separable data and are sensitive to outliers due to centroid initialization and mean computation.

Purpose of the Study:

• To propose an efficient multi-view K-Means algorithm that addresses the limitations of existing methods.
• To develop a clustering approach robust to outliers and capable of handling linearly inseparable data.

Main Methods:

• The proposed model avoids initialization and computation of cluster centroids.
• Butterworth filters transform adjacency matrices into distance matrices, enabling handling of linearly inseparable data.
• A tensor composed of discrete index matrices from different views is constructed, minimizing rank via tensor Schatten p-norm to leverage cross-view consistency and complementarity.

Main Results:

• Experiments on artificial datasets demonstrate the model's superiority on linearly inseparable data.
• Performance evaluations on benchmark datasets confirm the algorithm's effectiveness.

Conclusions:

• The developed multi-view K-Means algorithm offers an efficient and robust solution for complex, multi-source data clustering.
• The method's ability to handle non-separable data and outlier insensitivity marks a significant advancement in clustering techniques.