An ICA-based image fusion scheme using only source images

In this paper, we propose an image fusion scheme based on Independent Component Analysis (ICA), in which the ICA bases are trained from only source images. For traditional ICA-based image fusion algorithms, a number of sample images similar to source images are required to train the ICA bases. However, it is a difficult task to find such similar training images in practice. Since the image patches have intra-redundancy in one image and inter-similarity in source images, we select patches with representative characteristics of source images to train the ICA bases. Simulation results show that the proposed scheme preserves more information of source images and outperforms previous ICA-based image fusion methods.