BLIND SOURCE SEPARATION TECHNIQUE FOR LINEAR MIXTURES BASED ON ACCURATE ESTIMATION OF PROBABILITY DENSITY FUNCTION
Mohamed F. M. Fahmy, Usama Sayed Mohammed, Nahed Ali Bahran · JES. Journal of Engineering Sciences/JES. Journal of engineering sciences · 2010
This paper presents an accurate nonparametric method for evaluating signal's probability density function (pdf), as well as its entropy. It is based on using Bspline wavelets, as the smoothing filter for the data histogram distribution. Due to the excellent energy concentration feature of Bspline wavelets, this estimation was found to be accurate and robust of probability density function (pdf) estimation that are needed in some linear blind source separation (BSS) designs. The validity of the proposed technique is checked by its ability of recovering linearly blind source BSS, with a simple check to verify exact source recovery when no information is available about the mixing system. Several experiments have been carried out, to verify the ability of the proposed technique to accurately estimating signal's pdf as well as recovering linearly mixed signals and images with a simple independency check to determine whether exact separation is achieved or not.