Smoothly adjustable denoising using a priori knowledge

Assuming a signal is composed of information and noise, this paper presents a generic approach to denoising by mapping the noisy signal using a priori information about the signal to be retrieved. The method is based on a subspace decomposition of both the a priori information at disposal and the noisy signal, followed by shrinkage of both subspace coefficients and smooth mapping of first onto the second space. Our method propose 3 different ways of building the a priori knowledge: A model-base, averaging-based and recursive-based. The proposed method is particularly suited for low signal to noise ratio. The denoising methods are validated on synthetic electrocardiogram (ECG) signals and further assessed on real life ECG and visual brain evoked potentials using the wavelet transform.