Transformed Locally Linear Manifold Clustering

Jyoti Maggu (1) , Angshul Majumdar (1) , Emilie Chouzenoux (2, 3)

1 IIIT-Delhi - Indraprastha Institute of Information Technology [New Delhi]
2 GALEN-POST - Organ Modeling through Extraction, Representation and Understanding of Medical Image Content
3 LIGM - Laboratoire d'Informatique Gaspard-Monge

Transform learning is a relatively new analysis formulation for learning a basis to represent signals. This work incorporates the simplest subspace clustering formulation – Locally Linear Manifold Clustering, into the transform learning formulation. The core idea is to perform the clustering task in a transformed domain instead of processing directly the raw samples. The transform analysis step and the clustering are not done piecemeal but are performed jointly through the formulation of a coupled minimization problem. Comparison with state-of-the-art deep learning-based clustering methods and popular subspace clustering techniques shows that our formulation improves upon them.

Mots clés: Subspace clustering; Transform learning; Alternating optimization

Domaines: Traitement du signal et de l'image [eess.SP]

Jyoti Maggu, Angshul Majumdar, Emilie Chouzenoux. Transformed Locally Linear Manifold Clustering. EUSIPCO 2018 - 26th European Signal Processing Conference, Sep 2018, Rome, Italy. ⟨10.23919/EUSIPCO.2018.8553061⟩. ⟨hal-01862192⟩