Generalized Mixture Ratio Based Blind Image Separation

Abstract

Iterative approaches have been mostly used for blind source separation. The success of iterative methods generally depends upon appropriate selection of the values of the input parameters such as initial conditions and control parameters. We propose a direct non-iterative method for blindly separating linearly mixed images using the boundary values of the determinant of the relative transformation matrix between the mixtures. The algorithm is a non-trivial extension of the mixture ratio based algorithm originally proposed by Diamantaras and Papadimitriou (2005) and allows for a finite arbitrary number of source images. We demonstrate the performance of our method on synthetic images.