Clustering based dictionary learningCP6

Usual methods for dictionary learning applied to multidimensional data require a first vectorization step, loosing the intrinsic spatial correlation of the samples. We present a method based on a generalization of the Haar wavelet transform that builds a dictionary from a hierarchical clustering of the data. Depending on the chosen clustering method, no vectorization of the data is required. We will present numerical results in the case of two dimensional patches extracted from natural images.

This presentation is part of Contributed Presentation “CP6 - Contributed session 6

Renato Budinich (University of Göttingen)
Gerlind Plonka (University of Goettingen)
dictionary learning, image compression, image reconstruction, image representation, machine learning