Innovative ica and nmf methods for blind source separation of audio signals images and their performance comparison with existing methods

Abstract

Blind Signal Separation or Blind Source Separation (BSS) is taking away a set of source signals from a set of mixture signals or observation signals, without knowing any sort of information about the source signals or in some cases with only very little information about the source signals or about the mixing method. During the last couple of decades, BSS was acting as an emerging and interesting field in all forms of Signal Processing applications. It plays a significant role particularly in image processing and in audio signal processing applications. In healthcare applications normally if any scanned image or audio observation signal of a body part is considered, it may not be having information only about that particular part. It may include mixed information from nearby parts also. In such cases, with BSS processing the individual information is segregated out and with clear data the diagnosis is carried out. Formerly many different methods similar to Principal Component Analysis (PCA), Singular Value Decomposition (SVD), Independent Component Analysis (ICA), NonNegative Matrix Factorization (NMF), Dependent Component Analysis (DCA) and many more methods have been used. Each and every method has its own advantages and disadvantages. Among all the available methods ICA and NMF are in vogue. ICA is presently imperative and one of the oldest BSS method which can be applicable for any form of source signals for their separation. The chief advantage of ICA algorithms are, they can produce the separated sources with superior separation quality. The main drawback of this set of algorithms is their longer processing time for separation of the sources. Till now many different modifications were carried out on the conventional ICA method to set right this shortcoming where all those trials got failed in that aspect. Mutual Information based Least dependent Component Analysis algorithm (MILCA) is the latest form of ICA methods with the highest separation quality and dynamic behaviour among all other ICA methods. As other ICA group of algorithms this algorithm also has the same downside characteristic. But comparatively the MILCA algorithm takes very long duration for source separation among all ICA methods with relatively good separation quality. newline