ON THE USEOF KPCA TO EXTRACT ARTIFACTSINONE-DIMENSIONAL BIOMEDICALSIGNALS
Ana Rita Teixeira, Ana Maria Tomé, Deti Ieeta · 2006
Kernel principal component analysis(KPCA) isanonlinearprojective technique that canbeapplied todecompose multi-dimensional signals andextract informative features aswellasreduce anynoise contributions. Inthis workwe extend KPCAtoextract andremoveartifact-related contributions aswell asnoise fromone-dimensional signal recordings. We introduce anembedding stepwhichtransforms theone-dimensional signal intoamulti-dimensional vector.Thelatter isdecomposed infeature spacetoextract artifact related contaminations. Wefurther address thepreimageproblem andpropose aninitialization procedure to thefixed-point algorithm whichrenders itmoreefficient. Finally weapply KPCAtoextract dominant Electrooculogram(EOG)artifacts contaminating Electroencephalogram (EEG)recordings inafrontal channel.