Multi-sensor image fusion with ICA bases and region rule

Images captured from multi-sensor contain abundant information. It is available to combine them by image fusion schemes such as wavelet or Laplacian pyramid. Recently schemes based on independent component analysis (ICA) bases were proposed. These schemes transform images to ICA domain and then fuse them by fusion rules. Accordingly, the region based fusion rule segments images, selects features in transform domain and combines them. In this paper, we trained ICA bases which are suitable for image fusion and disposed of low frequency energy properly. To effectively make use of the relationship of input images, a novel region based fusion rule was proposed. In contrast to traditional schemes, the performance is improved and shown by both objective and subjective evaluations.