SPARSE BASED IMAGE FUSION USING COMPACT SUB-DICTIONARIES

Image fusion schemes are desirable to obtain a high-quality image by integrating complementary information from multiple source images. The main aim of this paper is to propose a novel image fusion technique that provides a highly informative image, which is useful in various applications like computer vision, medical diagnosis, remote sensing, etc. Traditional Sparse Representation (SR) based fusion method makes use of a single highly redundant dictionary for image fusion. This increases complexity and may also lead to visual artefacts in the fused image. Fusion scheme using dictionary-based sparse representation is proposed in this paper. A large number of image patches are pre-classified based on projections using R-Transform and a set of compact sub-dictionaries are learnt from them. At the fusion stage, one of the sub-dictionaries is chosen to fuse the given set of source images. Quantitative and Qualitative evaluation of the proposed fusion scheme on multi-focus and multi-modal images shows the superiority of the proposed scheme over other existing fusion algorithms.