Institute of Information Theory and Automation

You are here

Software for blind source separation and image sequence decomposition

This page contains algorithms published in different papers.

Bayesian non-negative matrix factorization with adaptive prior covariance (NMF-APC) algorithm

Toolbox with NMF-APC method submitted to IEEE Signal Processing Letters can be downloaded here.

 

Variational BSS Toolbox

Toolbox with algorithms presented in EUSIPCO 2015 conference. The code can be downloaded from here here.

 

  • Isotropic or sparse prior can be alter on both, source images and time-actvites. 

 

 

BSS with Various Prior on Weights Covariance Matrix

Algorithms presented in LVA/ICA 2015 conference. The code can be downloaded from here.

  • BSS algorithm with 5 variants of prior covariance matrix (isotropic, sparce, sparse differences, Wishart, Wishart with localization). 

 

Sparse Blind Source Separation and Deconvolution Using Sparsity Priors (S-BSS-vecDC)

Algorithm presented in IEEE Transaction on Medical Imaging 2015. See detail describtion here or download the algorithm from here. .

  • Blind source separation problem with convolution model using sparsity priors.
  • Both, source images and source convolution kernels are assumed to be sparse.

 

Blind Compartment Model Separation (BCMS)

Algorithm presented in VipIMAGE 2013 conference. The code can be downloaded from here.

  • Blind source separation with deconvolution while convolution kernels are modeled as piece-wise linear curves.

 

Sparse Blind Source Separation and Deconvolution (S-BSS-DC)

Algorithm presented in Šmídl Václav, Tichý Ondřej : Sparsity in Bayesian Blind Source Separation and Deconvolution , Machine Learning and Knowledge Discovery in Databases, p. 548-563, The European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML/PKDD 2013), (Praha, CZ, 24.09.2013-26.09.2013). The code can be downloaded from here.

Orthogonal Variational PCA

This code uses orthogonal probabilistic model of PCA to automatically determine the number of relevant principal components. See report for more details. The code can be downloaded from here.

Contact:

Support of grants

  • Image Blind Deconvolution in Demanding Conditions, Grant Agency of the Czech Republic, No. GA13-29225S, 2013-2016
  • Research center DAR, MŠMT 1M0572, 2005-2011

 

2019-03-14 11:51