Dictionary Learning with Low Mutual Coherence Constraint

This repository contains MATLAB functions and a demo script implementing and testing the low-coherence dictionary learning (DL) algorithms proposed in [1], namely, constrained incoherent DL (CINC-DL) and regularized incoherent DL (RINC-DL). We have also implemented the algorithms proposed in [2], bounded self-coherence DL (BSC-DL) as well as [3], iterative projections and rotations DL (IPR-DL), for comparison.

Copyright notice: We have largely used the the OMP and KSVD implementations written by Ron Rubinstein to write our codes.

Installation

Prior to use this package, you need to install the OMP and KSVD toolboxes. Then, to install the current package, simply run setup.m.

Package description

• DL_algorithms: MATLAB implementations of our proposed low-coherence dictionary learning algorithms, namely, RINC-DL and CINC-DL, together with two existing algorithms, namely, BSC-DL [2] and IPR-DL [3].

• Test_images: natural images used to test different low-coherence DL algorithms.

• demo.m: demo script to apply different low-coherence DL algorithms on natural image patches.

References

[1] M. Sadeghi and M. Babaie-Zadeh, Dictionary learning with low mutual coherence constraint, Neurocomputing, vol. 407, pp. 163-174, September 2020.

[2] C. D. Sigg, T. Dikk, and J. M. Buhmann, Learning dictionaries with bounded self-coherence, IEEE Signal Proc. Letters, vol. 19, no. 12, pp. 861-864, 2012.

[3] D. Barchiesi and M. D. Plumbley, Learning incoherent dictionaries for sparse approximation using iterative projections and rotations, IEEE Trans. on Signal Proc., vol. 61, no. 8, pp. 2055-2065, 2013.

Contact

Mostafa Sadeghi - mostafa[dot]sadeghi[at]inria[dot]fr