Research On The Noise Elimination Of Scaning Image Based On ICA

Independent Component Analysis is a kind of the signal processing theory, which is developed in the past two decades aimed to blind source separation. It is a methodology of statistics, which is designed to separate the independent source signals from the mixed signals received by the sensors so as to keep the separated source signals as independent as possible. It has wide application in the regard of voice recognition and signal processing concerning telecommunication, medicine, etc. Now it also has been a focus in the research field of artificial neural network and so on.The dissertation focuses on utilizing the ICA theory to remove unpleasant noise of the scanning images. When we scan documents, poor quality paper makes the words on rear side visible on front face; consequently, negative influence will be exerted on our normal reading. The disturbance information on the rear side is the noises which should be removed.In this research, the scanned images on both sides containing noises are processed as two mixed signals, which are separated to apply different ICA calculating methods i.e. Infomax, FastICA and the method based on generalized eigenvalue. In the different ICA methods, initial weighted value, nonlinear function and parameters can influent the experimental results to some extent. Therefore, many experiments are carried out by changing initial weighted value, nonlinear function and parameters for specific method. The experiment results have been analyzed in aspects of separation effect, processing time and algorithm stability and obtained corresponding feature of different methods regard to the practical problems.