Efficient Source Separation Enhancement Based on Advanced Multi-dimensional Transform Technique

Abstract

This paper is concerned with blind separation of digital images from mixtures. It suggests the implementation of a blind separation technique on the ridgelet transform (RT) of the mixed images, instead of executing the separation on the mixtures in the time or a trigonometric transform domain. Ridgelet transform is a new orientational multi-resolution transform, and it is widely appropriate for characterizing the signals with dimensional singularities. Finite ridgelet transform is a discrete implementation of the ridgelet transform, which has numerical accuracy as the uninterrupted RT and has soft calculation intricacy. In contrary to the time domain, the RT finds more applications in image separation, because it appears sleek and edge sides of images have sparsity. In addition, the RT involves additional orientation information. The separated images are obtained using independent component analysis. The simulation results reveal that image separation in the RT domain is better, when compared to separation in the time or trigonometric transform domains.