Publication

Abstract : Image fusion strives to generate a highly informational image by integrating the required data from various source images. Due to the superior capability of sparse representation to efficiently describe the structural information of an image, a robust dictionary based multi-sensor image fusion algorithm is developed. Initially, the raw images are first preprocessed with the top hat transform to improve contrast and visibility. The focus metrics for the patches of preprocessed source images are then computed. Patches with superior focus characteristics are chosen to build the over-complete dictionary. The sparse coefficients are computed once the learnt dictionary is acquired, and then the “max-L1” algorithm is performed to generate the fused coefficients. Finally, the learnt dictionary and fused sparse vectors are used to rebuild the fused image. The proposed fusion architecture provides improved visual quality and surpasses existing approaches quantitatively, according to experimental evidence.