IMAGE SEPARATION USING WAVELET TRANSFORM AND INDEDENDENT COMPONENT ANALYSIS

Abstract

Independent Component Analysis (ICA), which belongs to the general class of unsupervised learning algorithm, has been studied and applied for a few decades. One of the application field is Blind Source Separation (BSS), in which from the mixed signals we would like to recover the original source signal, whereas the mixing is unknown. In our work, the mixed images are preprocessed by the Discrete Wavelet Transform (DWT) multiresolution analysis, followed by the ICA estimation using InfoMax and FastICA algorithms. We do not contribute any new theoretical or algorithmic development. Instead, we carry out extensive simulations on many different types of images using various DWTs and ICA algorithms for comparison of their effectiveness.