Blind Separation Methods for Image Show-through Problem

This paper studies a image show-through problem. It happens often when we copy or scan a paper document, in which the image from the back page shows through. The images obtained on both side of the paper can be considered as mixture components, which are nonlinear mixtures of original images. In this study, we propose to use self-organizing map (SOM) and fastICA to implement separation of the image mixtures. SOM is neural network-based technique using unsupervised learning and can provide useful data representations. The separation results show that the two blind separation methods are applicable to the problem.