Denoising using local projective subspace methods

摘要: In this paper we present denoising algorithms for enhancing noisy signals based on Local ICA (LICA), Delayed AMUSE (dAMUSE) and Kernel PCA (KPCA). The algorithm LICA relies applying locally to clusters of embedded in a high-dimensional feature space delayed coordinates. components resembling the can be detected by various criteria like estimators kurtosis or variance autocorrelations depending statistical nature signal. proposed applied favorably problem multi-dimensional data. Another projective subspace method using coordinates has been recently with dAMUSE. It combines solution blind source separation problems efforts an elegant way proofs very efficient fast. Finally, KPCA represents non-linear that is well suited also. Besides illustrative applications toy examples images, provide application all considered analysis protein NMR spectra.