Blind Separation of Image Signals with Noise Detection and Estimation

We propose an independent component analysis (ICA) approach which is robust against impulse noise. It consists of noise detection and image signal separation. We introduce a self-organizing map (SOM) network to determine if the observed image pixels are corrupted by noise. We mark each pixel to distinguish normal and corrupted ones. After that, we use one of two traditional ICA algorithms (fixed-point algorithm and Gaussian moments-based fixed-point algorithm) to separate the images. The proposed approach has the capacity to recover the mixed images and reduce noise from observed images. The simulation results show that the proposed approach is suitable for practical unsupervised separation problem.